aicodeswitch 5.2.11 → 5.2.13

package/dist/server/proxy-server.js

@@ -52,7 +52,9 @@ const crypto_1 = __importDefault(require("crypto"));
 const streaming_1 = require("./transformers/streaming");
 const model_rewrite_transform_1 = require("./transformers/model-rewrite-transform");
 const chunk_collector_1 = require("./transformers/chunk-collector");
+const stream_timing_transform_1 = require("./transformers/stream-timing-transform");
 const rules_status_service_1 = require("./rules-status-service");
+const agent_map_1 = require("./agent-map");
 const index_1 = require("./conversions/index");
 const stream_converter_adapter_1 = require("./conversions/stream-converter-adapter");
 const types_1 = require("../types");
@@ -219,6 +221,12 @@ class ProxyServer {
             writable: true,
             value: null
         });
+        Object.defineProperty(this, "performanceTracker", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: null
+        });
         // 请求去重缓存：用于防止同一个请求被重复计数（如网络重试）
         // key: requestHash, value: timestamp
         Object.defineProperty(this, "requestDedupeCache", {
@@ -253,6 +261,43 @@ class ProxyServer {
     getAccessKeyModule() {
         return this.accessKeyModule;
     }
+    /** 设置服务性能统计 tracker（全局，与 AUTH 无关） */
+    setPerformanceTracker(tracker) {
+        this.performanceTracker = tracker;
+    }
+    /** 获取服务性能统计 tracker */
+    getPerformanceTracker() {
+        return this.performanceTracker;
+    }
+    /**
+     * 采集一次请求的服务性能数据点（全局，与 AUTH 无关）。
+     * 在两条转发路径的 finalizeLog 公共点调用，覆盖 AccessKey + 普通路由。
+     * 流式：依据 streamTiming 精确计算 TTFT 与生成阶段吞吐；非流式：端到端估算（estimated）。
+     */
+    emitPerformance(params) {
+        const tracker = this.performanceTracker;
+        if (!tracker)
+            return;
+        const { statusCode, startTime, usage, streamTiming, service, vendorId, vendorName, model } = params;
+        const isError = statusCode >= 400;
+        const outputTokens = usage === null || usage === void 0 ? void 0 : usage.outputTokens;
+        const responseMs = Date.now() - startTime;
+        let ttftMs;
+        let tokensPerSecond;
+        let timingAccuracy = 'estimated';
+        if (streamTiming && streamTiming.hasTiming()) {
+            timingAccuracy = 'precise';
+            ttftMs = streamTiming.firstEventAt - startTime;
+            const generationMs = streamTiming.lastEventAt - streamTiming.firstEventAt;
+            if (outputTokens && generationMs > 0) {
+                tokensPerSecond = outputTokens / (generationMs / 1000);
+            }
+        }
+        else if (outputTokens && responseMs > 0) {
+            tokensPerSecond = outputTokens / (responseMs / 1000);
+        }
+        tracker.recordPerformance(vendorId !== null && vendorId !== void 0 ? vendorId : service.vendorId, vendorName, service.id, service.name, model, { ttftMs, tokensPerSecond, outputTokens, timingAccuracy, isError });
+    }
     /**
      * 从请求中提取 API Key（支持三种 Header，按优先级依次尝试）
      */
@@ -705,6 +750,7 @@ class ProxyServer {
                     responseTime: Date.now() - requestStartAt,
                     // 添加最后失败的服务信息
                     ruleId: lastFailedRule === null || lastFailedRule === void 0 ? void 0 : lastFailedRule.id,
+                    routeId: route === null || route === void 0 ? void 0 : route.id,
                     targetServiceId: lastFailedService === null || lastFailedService === void 0 ? void 0 : lastFailedService.id,
                     targetServiceName: lastFailedService === null || lastFailedService === void 0 ? void 0 : lastFailedService.name,
                     targetModel: (lastFailedRule === null || lastFailedRule === void 0 ? void 0 : lastFailedRule.targetModel) || ((_e = req.body) === null || _e === void 0 ? void 0 : _e.model),
@@ -3371,7 +3417,7 @@ class ProxyServer {
     }
     proxyRequest(req, res, route, rule, service, options) {
         return __awaiter(this, void 0, void 0, function* () {
-            var _a, _b, _c, _d, _e, _f, _g, _h, _j, _k, _l, _m, _o, _p, _q, _r, _s;
+            var _a, _b, _c, _d, _e, _f, _g, _h, _j, _k, _l, _m, _o, _p, _q, _r, _s, _t, _u;
             res.locals.skipLog = true;
             const startTime = Date.now();
             const rawSourceType = service.sourceType || 'openai-chat';
@@ -3379,8 +3425,15 @@ class ProxyServer {
             const sourceType = (0, type_migration_1.normalizeSourceType)(rawSourceType);
             const targetType = this.inferToolFromRequest(req);
             const sessionId = this.defaultExtractSessionId(req, targetType) || '-';
+            // Agent Map：在途请求注册（active 状态判定依据）
+            const _accessKeyCtxAtStart = req._accessKeyCtx;
+            agent_map_1.agentMapService.startRequest(sessionId, targetType, {
+                source: _accessKeyCtxAtStart ? 'access-key' : 'global',
+                keyId: (_a = _accessKeyCtxAtStart === null || _accessKeyCtxAtStart === void 0 ? void 0 : _accessKeyCtxAtStart.accessKey) === null || _a === void 0 ? void 0 : _a.id,
+                keyName: (_b = _accessKeyCtxAtStart === null || _accessKeyCtxAtStart === void 0 ? void 0 : _accessKeyCtxAtStart.accessKey) === null || _b === void 0 ? void 0 : _b.name,
+            });
             const vendor = this.dbManager.getVendorByServiceId(service.id);
-            console.log(`\x1b[32m[Request Start]\x1b[0m client=${targetType}, session=${sessionId}, rule=${rule.id}(${rule.contentType}), vendor=${(vendor === null || vendor === void 0 ? void 0 : vendor.name) || '-'}, service=${service.name}, model=${rule.targetModel || ((_a = req.body) === null || _a === void 0 ? void 0 : _a.model) || '-'}`);
+            console.log(`\x1b[32m[Request Start]\x1b[0m client=${targetType}, session=${sessionId}, rule=${rule.id}(${rule.contentType}), vendor=${(vendor === null || vendor === void 0 ? void 0 : vendor.name) || '-'}, service=${service.name}, model=${rule.targetModel || ((_c = req.body) === null || _c === void 0 ? void 0 : _c.model) || '-'}`);
             const failoverEnabled = (options === null || options === void 0 ? void 0 : options.failoverEnabled) === true;
             const forwardedToServiceName = options === null || options === void 0 ? void 0 : options.forwardedToServiceName;
             const useOriginalConfig = (options === null || options === void 0 ? void 0 : options.useOriginalConfig) === true;
@@ -3544,12 +3597,45 @@ class ProxyServer {
             let downstreamResponseBodyForLog;
             let upstreamRequestForLog;
             let actuallyUsedProxy = false; // 标记是否实际使用了代理
+            // 服务性能打点：流式分支会创建实例并注入 pipeline；finalizeLog 据此判定 precise/estimated
+            let streamTiming = null;
             // 标记规则正在使用
             rules_status_service_1.rulesStatusBroadcaster.markRuleInUse(route.id, rule.id);
             const finalizeLog = (statusCode, error) => __awaiter(this, void 0, void 0, function* () {
-                var _a, _b, _c, _d, _e, _f, _g;
+                var _a, _b, _c, _d, _e, _f, _g, _h, _j, _k, _l, _m, _o;
                 if (logged)
                     return;
+                // 服务性能数据点采集（全局，与 AUTH 无关；独立于 enableLogging 开关）
+                this.emitPerformance({
+                    statusCode, startTime, usage: usageForLog, streamTiming,
+                    service, vendorId: vendor === null || vendor === void 0 ? void 0 : vendor.id, vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
+                    model: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_a = req.body) === null || _a === void 0 ? void 0 : _a.model),
+                });
+                // Agent Map：在途请求注销 + 活动/状态采集广播（独立于 enableLogging，仅执行一次）
+                if (!res.locals._agentMapRecorded) {
+                    res.locals._agentMapRecorded = true;
+                    agent_map_1.agentMapService.endRequest(sessionId);
+                    const _akCtx = req._accessKeyCtx;
+                    const _tokensDelta = (usageForLog === null || usageForLog === void 0 ? void 0 : usageForLog.totalTokens) ||
+                        (((usageForLog === null || usageForLog === void 0 ? void 0 : usageForLog.inputTokens) || 0) + ((usageForLog === null || usageForLog === void 0 ? void 0 : usageForLog.outputTokens) || 0));
+                    agent_map_1.agentMapService.onFinalized({
+                        sessionId,
+                        agent: targetType,
+                        source: _akCtx ? 'access-key' : 'global',
+                        keyId: (_b = _akCtx === null || _akCtx === void 0 ? void 0 : _akCtx.accessKey) === null || _b === void 0 ? void 0 : _b.id,
+                        keyName: (_c = _akCtx === null || _akCtx === void 0 ? void 0 : _akCtx.accessKey) === null || _c === void 0 ? void 0 : _c.name,
+                        title: this.defaultExtractSessionTitle(req, sessionId),
+                        timestamp: Date.now(),
+                        statusCode,
+                        // 采用真正转发给上游的模型（转换+覆盖后的 requestBody.model），而非 rule.targetModel 预测值，
+                        // 避免规则未配置 targetModel 时回退成编程工具提交的 req.body.model
+                        model: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_d = req.body) === null || _d === void 0 ? void 0 : _d.model),
+                        tokensDelta: _tokensDelta,
+                        body: req.body,
+                        downstreamResponseBody: downstreamResponseBodyForLog !== null && downstreamResponseBodyForLog !== void 0 ? downstreamResponseBodyForLog : responseBodyForLog,
+                        responseBody: responseBodyForLog,
+                    });
+                }
                 const isError = statusCode >= 400;
                 if (isError) {
                     console.log(`\x1b[31m[Request Error]\x1b[0m client=${targetType}, session=${sessionId}, rule=${rule.id}(${rule.contentType}), vendor=${(vendor === null || vendor === void 0 ? void 0 : vendor.name) || '-'}, service=${service.name}, status=${statusCode}, time=${Date.now() - startTime}ms${error ? `, error=${error}` : ''}`);
@@ -3558,7 +3644,7 @@ class ProxyServer {
                     console.log(`\x1b[33m[Request End]\x1b[0m client=${targetType}, session=${sessionId}, rule=${rule.id}(${rule.contentType}), vendor=${(vendor === null || vendor === void 0 ? void 0 : vendor.name) || '-'}, service=${service.name}, status=${statusCode}, time=${Date.now() - startTime}ms`);
                 }
                 // 检查是否启用日志记录（默认启用）
-                const enableLogging = ((_a = this.config) === null || _a === void 0 ? void 0 : _a.enableLogging) !== false; // 默认为 true
+                const enableLogging = ((_e = this.config) === null || _e === void 0 ? void 0 : _e.enableLogging) !== false; // 默认为 true
                 if (!enableLogging) {
                     return;
                 }
@@ -3582,13 +3668,14 @@ class ProxyServer {
                             error,
                             contentType: rule.contentType,
                             ruleId: rule.id,
+                            routeId: route === null || route === void 0 ? void 0 : route.id,
                             targetType,
                             targetServiceId: service.id,
                             targetServiceName: service.name,
-                            targetModel: rule.targetModel || ((_b = req.body) === null || _b === void 0 ? void 0 : _b.model),
+                            targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_f = req.body) === null || _f === void 0 ? void 0 : _f.model),
                             vendorId: service.vendorId,
                             vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                            requestModel: (_c = req.body) === null || _c === void 0 ? void 0 : _c.model,
+                            requestModel: (_g = req.body) === null || _g === void 0 ? void 0 : _g.model,
                             tags: this.buildRelayTags(relayedForLog, useOriginalConfig),
                             responseHeaders: responseHeadersForLog,
                             responseBody: responseBodyForLog,
@@ -3622,7 +3709,7 @@ class ProxyServer {
                                 vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
                                 serviceId: service.id,
                                 serviceName: service.name,
-                                model: rule.targetModel || ((_d = req.body) === null || _d === void 0 ? void 0 : _d.model),
+                                model: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_h = req.body) === null || _h === void 0 ? void 0 : _h.model),
                                 totalTokens: sessionTokens,
                             }).catch(err => console.error('[KeySession] upsert error:', err));
                         }
@@ -3646,13 +3733,14 @@ class ProxyServer {
                             error,
                             contentType: rule.contentType,
                             ruleId: rule.id,
+                            routeId: route === null || route === void 0 ? void 0 : route.id,
                             targetType,
                             targetServiceId: service.id,
                             targetServiceName: service.name,
-                            targetModel: rule.targetModel || ((_e = req.body) === null || _e === void 0 ? void 0 : _e.model),
+                            targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_j = req.body) === null || _j === void 0 ? void 0 : _j.model),
                             vendorId: service.vendorId,
                             vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                            requestModel: (_f = req.body) === null || _f === void 0 ? void 0 : _f.model,
+                            requestModel: (_k = req.body) === null || _k === void 0 ? void 0 : _k.model,
                             tags: this.buildRelayTags(relayedForLog, useOriginalConfig),
                         });
                     }
@@ -3665,7 +3753,7 @@ class ProxyServer {
                 const vendors = this.dbManager.getVendors();
                 const vendorForLog = vendors.find(v => v.id === service.vendorId);
                 // 从请求体中提取模型信息
-                const requestModel = (_g = req.body) === null || _g === void 0 ? void 0 : _g.model;
+                const requestModel = (_l = req.body) === null || _l === void 0 ? void 0 : _l.model;
                 const tagsForLog = this.buildRelayTags(relayedForLog, useOriginalConfig);
                 if (extraTagsForLog.length > 0) {
                     tagsForLog.push(...extraTagsForLog);
@@ -3684,10 +3772,11 @@ class ProxyServer {
                     // 新增字段
                     contentType: rule.contentType,
                     ruleId: rule.id,
+                    routeId: route === null || route === void 0 ? void 0 : route.id,
                     targetType,
                     targetServiceId: service.id,
                     targetServiceName: service.name,
-                    targetModel: rule.targetModel || requestModel,
+                    targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_m = req.body) === null || _m === void 0 ? void 0 : _m.model),
                     vendorId: service.vendorId,
                     vendorName: vendorForLog === null || vendorForLog === void 0 ? void 0 : vendorForLog.name,
                     requestModel,
@@ -3715,7 +3804,7 @@ class ProxyServer {
                         vendorName: vendorForLog === null || vendorForLog === void 0 ? void 0 : vendorForLog.name,
                         serviceId: service.id,
                         serviceName: service.name,
-                        model: requestModel || rule.targetModel,
+                        model: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_o = req.body) === null || _o === void 0 ? void 0 : _o.model),
                         totalTokens,
                         highIqMode: rule.contentType === 'high-iq' ? true : existingSession === null || existingSession === void 0 ? void 0 : existingSession.highIqMode,
                         highIqRuleId: rule.contentType === 'high-iq' ? rule.id : existingSession === null || existingSession === void 0 ? void 0 : existingSession.highIqRuleId,
@@ -3816,7 +3905,7 @@ class ProxyServer {
                     targetType,
                     targetServiceId: service.id,
                     targetServiceName: service.name,
-                    targetModel: rule.targetModel || ((_a = req.body) === null || _a === void 0 ? void 0 : _a.model),
+                    targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_a = req.body) === null || _a === void 0 ? void 0 : _a.model),
                     vendorId: service.vendorId,
                     vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
                     requestModel: (_b = req.body) === null || _b === void 0 ? void 0 : _b.model,
@@ -3882,7 +3971,7 @@ class ProxyServer {
                 // responses→responses 直连非 OpenAI 官方端点时，需降级兼容（剥离 custom/namespace 等私有工具与非标准字段）
                 const sanitizeBody = clientFormat === 'responses' && (0, source_type_mapping_1.sourceTypeToFormat)(sourceType) === 'responses' && !(0, index_1.isOfficialOpenAiApi)(effectiveApiUrl || '');
                 const transformedRequestBody = this.transformRequestToUpstream(targetType, sourceType, payloadForTransform, rule.targetModel, providerConfig, serverToolConfig, sanitizeBody);
-                requestBody = (_b = transformedRequestBody !== null && transformedRequestBody !== void 0 ? transformedRequestBody : this.cloneRequestBody(originalToolRequestBody)) !== null && _b !== void 0 ? _b : {};
+                requestBody = (_d = transformedRequestBody !== null && transformedRequestBody !== void 0 ? transformedRequestBody : this.cloneRequestBody(originalToolRequestBody)) !== null && _d !== void 0 ? _d : {};
                 // 对最终即将发送到上游的 Claude compact 请求再做一次兜底清理，
                 // 避免中间转换/覆盖步骤重新引入未配对的 tool_use。
                 if (rule.contentType === 'compact' && targetType === 'claude-code' && Array.isArray(requestBody === null || requestBody === void 0 ? void 0 : requestBody.messages)) {
@@ -3988,7 +4077,7 @@ class ProxyServer {
                     let errorResponseData = response.data;
                     if (streamRequested && response.data && typeof response.data.on === 'function') {
                         const raw = yield this.readStreamBody(response.data);
-                        errorResponseData = (_c = this.safeJsonParse(raw)) !== null && _c !== void 0 ? _c : raw;
+                        errorResponseData = (_e = this.safeJsonParse(raw)) !== null && _e !== void 0 ? _e : raw;
                     }
                     responseHeadersForLog = this.normalizeResponseHeaders(responseHeaders);
                     yield handleUpstreamHttpError(response.status, errorResponseData, responseHeaders, contentType);
@@ -4026,10 +4115,10 @@ class ProxyServer {
                                 targetType,
                                 targetServiceId: service.id,
                                 targetServiceName: service.name,
-                                targetModel: rule.targetModel || ((_d = req.body) === null || _d === void 0 ? void 0 : _d.model),
+                                targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_f = req.body) === null || _f === void 0 ? void 0 : _f.model),
                                 vendorId: service.vendorId,
                                 vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                                requestModel: (_e = req.body) === null || _e === void 0 ? void 0 : _e.model,
+                                requestModel: (_g = req.body) === null || _g === void 0 ? void 0 : _g.model,
                                 responseTime: Date.now() - startTime,
                             });
                         }
@@ -4061,11 +4150,13 @@ class ProxyServer {
                     const downstreamChunkCollector = new chunk_collector_1.ChunkCollectorTransform(() => {
                         rules_status_service_1.rulesStatusBroadcaster.refreshRuleInUse(route.id, rule.id);
                     });
+                    // 服务性能打点：记录首/末 SSE 事件时间，用于 TTFT 与生成阶段吞吐
+                    streamTiming = new stream_timing_transform_1.StreamTimingTransform(startTime);
                     const compactResponseSanitizer = rule.contentType === 'compact' && targetType === 'claude-code'
                         ? new ClaudeCompactResponseSanitizer()
                         : null;
                     // 流式 model 回写：将上游返回的 model 改写为客户端请求时的原始模型名
-                    const originalModel = (_f = req.body) === null || _f === void 0 ? void 0 : _f.model;
+                    const originalModel = (_h = req.body) === null || _h === void 0 ? void 0 : _h.model;
                     const modelRewriter = originalModel ? new model_rewrite_transform_1.ModelRewriteTransform(originalModel) : null;
                     responseHeadersForLog = this.normalizeResponseHeaders(responseHeaders);
                     // 使用 transformSSEToTool 方法选择转换器
@@ -4102,7 +4193,7 @@ class ProxyServer {
                         ensureResponseWritable();
                         return yield new Promise((resolve, reject) => {
                             if (converter) {
-                                const streamStages = [streamSource, parser, eventCollector, converter];
+                                const streamStages = [streamSource, parser, eventCollector, streamTiming, converter];
                                 if (compactResponseSanitizer) {
                                     streamStages.push(compactResponseSanitizer);
                                 }
@@ -4120,7 +4211,7 @@ class ProxyServer {
                                 });
                                 return;
                             }
-                            const streamStages = [streamSource, parser, eventCollector];
+                            const streamStages = [streamSource, parser, eventCollector, streamTiming];
                             if (compactResponseSanitizer) {
                                 streamStages.push(compactResponseSanitizer);
                             }
@@ -4165,10 +4256,10 @@ class ProxyServer {
                                 targetType,
                                 targetServiceId: service.id,
                                 targetServiceName: service.name,
-                                targetModel: rule.targetModel || ((_g = req.body) === null || _g === void 0 ? void 0 : _g.model),
+                                targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_j = req.body) === null || _j === void 0 ? void 0 : _j.model),
                                 vendorId: service.vendorId,
                                 vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                                requestModel: (_h = req.body) === null || _h === void 0 ? void 0 : _h.model,
+                                requestModel: (_k = req.body) === null || _k === void 0 ? void 0 : _k.model,
                                 responseTime: Date.now() - startTime,
                             });
                         }
@@ -4212,10 +4303,10 @@ class ProxyServer {
                                 targetType,
                                 targetServiceId: service.id,
                                 targetServiceName: service.name,
-                                targetModel: rule.targetModel || ((_j = req.body) === null || _j === void 0 ? void 0 : _j.model),
+                                targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_l = req.body) === null || _l === void 0 ? void 0 : _l.model),
                                 vendorId: service.vendorId,
                                 vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                                requestModel: (_k = req.body) === null || _k === void 0 ? void 0 : _k.model,
+                                requestModel: (_m = req.body) === null || _m === void 0 ? void 0 : _m.model,
                                 responseTime: Date.now() - startTime,
                             });
                         }
@@ -4234,7 +4325,7 @@ class ProxyServer {
                 let responseData = response.data;
                 if (streamRequested && response.data && typeof response.data.on === 'function' && !isEventStream) {
                     const raw = yield this.readStreamBody(response.data);
-                    responseData = (_l = this.safeJsonParse(raw)) !== null && _l !== void 0 ? _l : raw;
+                    responseData = (_o = this.safeJsonParse(raw)) !== null && _o !== void 0 ? _o : raw;
                 }
                 // 收集响应头
                 responseHeadersForLog = this.normalizeResponseHeaders(responseHeaders);
@@ -4256,7 +4347,7 @@ class ProxyServer {
                 usageForLog = this.extractTokenUsageFromResponse(responseData, sourceType);
                 console.log('[Proxy] Non-stream response: extracted usageForLog:', usageForLog);
                 // 回写 model 字段：将上游返回的 model 改写为客户端请求时的原始模型名
-                const originalModel = (_m = req.body) === null || _m === void 0 ? void 0 : _m.model;
+                const originalModel = (_p = req.body) === null || _p === void 0 ? void 0 : _p.model;
                 (0, model_rewrite_transform_1.rewriteResponseModel)(normalizedConverted, originalModel);
                 (0, model_rewrite_transform_1.rewriteResponseModel)(responseData, originalModel);
                 this.copyResponseHeaders(responseHeaders, res);
@@ -4321,10 +4412,10 @@ class ProxyServer {
                         targetType,
                         targetServiceId: service.id,
                         targetServiceName: service.name,
-                        targetModel: rule.targetModel || ((_o = req.body) === null || _o === void 0 ? void 0 : _o.model),
+                        targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_q = req.body) === null || _q === void 0 ? void 0 : _q.model),
                         vendorId: service.vendorId,
                         vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                        requestModel: (_p = req.body) === null || _p === void 0 ? void 0 : _p.model,
+                        requestModel: (_r = req.body) === null || _r === void 0 ? void 0 : _r.model,
                         upstreamRequest: upstreamRequestForLog,
                         responseTime: Date.now() - startTime,
                     });
@@ -4339,7 +4430,7 @@ class ProxyServer {
                 console.error('Proxy error:', error);
                 // 检测是否是 timeout 错误
                 const isTimeout = error.code === 'ECONNABORTED' ||
-                    ((_q = error.message) === null || _q === void 0 ? void 0 : _q.toLowerCase().includes('timeout')) ||
+                    ((_s = error.message) === null || _s === void 0 ? void 0 : _s.toLowerCase().includes('timeout')) ||
                     (error.errno && error.errno === 'ETIMEDOUT');
                 const statusCode = isTimeout ? 504 : this.getErrorStatusCode(error, 500);
                 const baseErrorMessage = isTimeout
@@ -4365,10 +4456,10 @@ class ProxyServer {
                     targetType,
                     targetServiceId: service.id,
                     targetServiceName: service.name,
-                    targetModel: rule.targetModel || ((_r = req.body) === null || _r === void 0 ? void 0 : _r.model),
+                    targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_t = req.body) === null || _t === void 0 ? void 0 : _t.model),
                     vendorId: service.vendorId,
                     vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                    requestModel: (_s = req.body) === null || _s === void 0 ? void 0 : _s.model,
+                    requestModel: (_u = req.body) === null || _u === void 0 ? void 0 : _u.model,
                     upstreamRequest: upstreamRequestForLog,
                     responseHeaders: responseHeadersForLog,
                     responseTime: Date.now() - startTime,
@@ -4608,6 +4699,8 @@ class ProxyServer {
             let responseBodyForLog;
             let downstreamResponseBodyForLog;
             let streamChunksForLog;
+            // 服务性能打点：流式分支会创建实例并注入 pipeline
+            let streamTiming = null;
             let responseHeadersForLog;
             let upstreamRequestForLog;
             let relayedForLog = true;
@@ -4629,10 +4722,16 @@ class ProxyServer {
                 requestBody = (0, compact_1.normalizeClaudeCompactRequestBody)(requestBody);
             }
             const finalizeLog = (statusCode, error) => __awaiter(this, void 0, void 0, function* () {
-                var _a, _b, _c, _d, _e;
+                var _a, _b, _c, _d, _e, _f;
                 if (logged)
                     return;
                 logged = true;
+                // 服务性能数据点采集（全局，与 AUTH 无关；独立于 enableLogging 开关）
+                this.emitPerformance({
+                    statusCode, startTime, usage: usageForLog, streamTiming,
+                    service, vendorId: vendor === null || vendor === void 0 ? void 0 : vendor.id, vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
+                    model: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_a = req.body) === null || _a === void 0 ? void 0 : _a.model),
+                });
                 // AccessKey 独立日志处理
                 const accessKeyCtx = req._accessKeyCtx;
                 if (accessKeyCtx && this.accessKeyModule) {
@@ -4649,12 +4748,13 @@ class ProxyServer {
                             error,
                             contentType: rule.contentType,
                             ruleId: rule.id,
+                            routeId: route === null || route === void 0 ? void 0 : route.id,
                             targetServiceId: service.id,
                             targetServiceName: service.name,
-                            targetModel: rule.targetModel || ((_a = req.body) === null || _a === void 0 ? void 0 : _a.model),
+                            targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_b = req.body) === null || _b === void 0 ? void 0 : _b.model),
                             vendorId: service.vendorId,
                             vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                            requestModel: (_b = req.body) === null || _b === void 0 ? void 0 : _b.model,
+                            requestModel: (_c = req.body) === null || _c === void 0 ? void 0 : _c.model,
                             tags: this.buildRelayTags(relayedForLog),
                         });
                         if (usageForLog && statusCode < 400) {
@@ -4680,7 +4780,7 @@ class ProxyServer {
                                 vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
                                 serviceId: service.id,
                                 serviceName: service.name,
-                                model: rule.targetModel || ((_c = req.body) === null || _c === void 0 ? void 0 : _c.model),
+                                model: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_d = req.body) === null || _d === void 0 ? void 0 : _d.model),
                                 totalTokens: sessionTokens,
                             }).catch(err => console.error('[KeySession] upsert error:', err));
                         }
@@ -4702,12 +4802,13 @@ class ProxyServer {
                             error,
                             contentType: rule.contentType,
                             ruleId: rule.id,
+                            routeId: route === null || route === void 0 ? void 0 : route.id,
                             targetServiceId: service.id,
                             targetServiceName: service.name,
-                            targetModel: rule.targetModel || ((_d = req.body) === null || _d === void 0 ? void 0 : _d.model),
+                            targetModel: (requestBody === null || requestBody === void 0 ? void 0 : requestBody.model) || ((_e = req.body) === null || _e === void 0 ? void 0 : _e.model),
                             vendorId: service.vendorId,
                             vendorName: vendor === null || vendor === void 0 ? void 0 : vendor.name,
-                            requestModel: (_e = req.body) === null || _e === void 0 ? void 0 : _e.model,
+                            requestModel: (_f = req.body) === null || _f === void 0 ? void 0 : _f.model,
                             tags: this.buildRelayTags(relayedForLog),
                         });
                     }
@@ -4877,6 +4978,8 @@ class ProxyServer {
                     const downstreamChunkCollector = new chunk_collector_1.ChunkCollectorTransform(() => {
                         rules_status_service_1.rulesStatusBroadcaster.refreshRuleInUse(route.id, rule.id);
                     });
+                    // 服务性能打点：记录首/末 SSE 事件时间
+                    streamTiming = new stream_timing_transform_1.StreamTimingTransform(startTime);
                     responseHeadersForLog = this.normalizeResponseHeaders(responseHeaders);
                     // 流式 model 回写：将上游返回的 model 改写为客户端请求时的原始模型名
                     const originalModel = (_d = req.body) === null || _d === void 0 ? void 0 : _d.model;
@@ -4911,7 +5014,7 @@ class ProxyServer {
                                 return stages;
                             };
                             if (converter) {
-                                const stages = buildStages(streamSource, parser, eventCollector, converter);
+                                const stages = buildStages(streamSource, parser, eventCollector, streamTiming, converter);
                                 stream_1.pipeline(...stages, (error) => {
                                     if (error) {
                                         reject(error);
@@ -4921,7 +5024,7 @@ class ProxyServer {
                                 });
                             }
                             else {
-                                const stages = buildStages(streamSource, parser, eventCollector);
+                                const stages = buildStages(streamSource, parser, eventCollector, streamTiming);
                                 stream_1.pipeline(...stages, (error) => {
                                     if (error) {
                                         reject(error);

package/dist/server/rules-status-service.js

@@ -52,8 +52,8 @@ class RulesStatusBroadcaster extends events_1.EventEmitter {
             enumerable: true,
             configurable: true,
             writable: true,
-            value: 10000
-        }); // 10秒无活动后标记为空闲
+            value: 120000
+        }); // 120秒无活动后标记为空闲（兜底安全网，覆盖 thinking hold 等长静默场景）
         Object.defineProperty(this, "IDLE_DEBOUNCE_DELAY", {
             enumerable: true,
             configurable: true,
@@ -258,16 +258,31 @@ class RulesStatusBroadcaster extends events_1.EventEmitter {
         });
     }
     /**
-     * 刷新规则使用中的不活动定时器（轻量级，仅重置定时器，不修改状态）
-     * 用于 streaming 过程中持续保持 in_use 状态
+     * 刷新规则使用中的不活动定时器（轻量级，仅重置定时器，通常不修改状态）
+     * 用于 streaming 过程中持续保持 in_use 状态。
+     *
+     * 行为：
+     * - status === 'in_use'：重置不活动定时器，并清除可能 pending 的 idle debounce，
+     *   避免已触发的 idle 经 SSE 推送出去（thinking hold 场景的关键修复）。
+     * - status === 'idle'：说明此前已被错误判空闲，但请求仍在出流——重新标记为 in_use
+     *   以便经 SSE 把状态推回"使用中"，实现前端自愈。
+     * - status === 'error' / 'suspended'：早退，这两种终态有独立恢复机制，不应被流式刷新覆盖。
      */
     refreshRuleInUse(routeId, ruleId) {
         const currentStatus = this.ruleStates.get(ruleId);
-        // 仅当状态已经是 in_use 时才刷新定时器
-        if ((currentStatus === null || currentStatus === void 0 ? void 0 : currentStatus.status) !== 'in_use')
+        // 终态有独立恢复机制，刷新不应覆盖
+        if ((currentStatus === null || currentStatus === void 0 ? void 0 : currentStatus.status) === 'error' || (currentStatus === null || currentStatus === void 0 ? void 0 : currentStatus.status) === 'suspended') {
             return;
+        }
         const timeoutKey = `${routeId}:${ruleId}`;
+        // 已被错误判空闲：重新标记为 in_use（内部会清旧定时器/debounce 并 emit statusChanged → SSE 推回使用中）
+        if ((currentStatus === null || currentStatus === void 0 ? void 0 : currentStatus.status) === 'idle') {
+            this.markRuleInUse(routeId, ruleId);
+            return;
+        }
+        // in_use：重置不活动定时器，并清除 pending 的 idle debounce（阻止已触发的 idle 经 SSE 推送）
         this.clearRuleTimeout(timeoutKey);
+        this.clearIdleDebounce(timeoutKey);
         const timeout = setTimeout(() => {
             this.markRuleIdle(routeId, ruleId);
         }, this.INACTIVITY_TIMEOUT);

package/dist/server/transformers/stream-timing-transform.js

@@ -0,0 +1,63 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.StreamTimingTransform = void 0;
+const stream_1 = require("stream");
+/**
+ * StreamTimingTransform - 流式打点 Transform（服务性能统计专用）
+ *
+ * 透传所有数据（对象模式 / Buffer / string 均可），仅记录：
+ *   - firstEventAt：首个被解析出的 SSE 事件流经本 Transform 的时刻（≈ 首 Token 返回时刻）
+ *   - lastEventAt：最后一个事件流经的时刻（≈ 整个返回结束时刻）
+ *
+ * 由调用方在转发开始前注入 startTime（请求发起时刻），即可派生：
+ *   - ttftMs       = firstEventAt - startTime
+ *   - generationMs = lastEventAt - firstEventAt
+ *
+ * 该 Transform 仅做时间记录，不修改任何数据内容，对转发链路零影响。
+ */
+class StreamTimingTransform extends stream_1.Transform {
+    constructor(startTime) {
+        super({ writableObjectMode: true, readableObjectMode: true });
+        /** 请求发起时刻（由外部注入） */
+        Object.defineProperty(this, "startTime", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        /** 首个事件到达时刻（未收到则为 0） */
+        Object.defineProperty(this, "firstEventAt", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: 0
+        });
+        /** 最后一个事件到达时刻 */
+        Object.defineProperty(this, "lastEventAt", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: 0
+        });
+        this.startTime = startTime;
+    }
+    _transform(chunk, _encoding, callback) {
+        try {
+            const now = Date.now();
+            if (this.firstEventAt === 0) {
+                this.firstEventAt = now;
+            }
+            this.lastEventAt = now;
+            this.push(chunk);
+        }
+        catch (error) {
+            console.error('[StreamTimingTransform] Error in _transform:', error);
+        }
+        callback();
+    }
+    /** 是否采集到至少一个事件（用于判定精确口径可用性） */
+    hasTiming() {
+        return this.firstEventAt > 0 && this.lastEventAt > 0;
+    }
+}
+exports.StreamTimingTransform = StreamTimingTransform;