chore: sync VERSION file with release v1.1.241 [skip ci]

feat(admin): 添加账户导出同步 API

.env.example

@@ -61,6 +61,9 @@ PROXY_USE_IPV4=true
 # ⏱️ 请求超时配置
 REQUEST_TIMEOUT=600000  # 请求超时设置（毫秒），默认10分钟
 
+# 🔧 请求体大小配置
+REQUEST_MAX_SIZE_MB=60
+
 # 📈 使用限制
 DEFAULT_TOKEN_LIMIT=1000000
 
@@ -75,6 +78,8 @@ TOKEN_USAGE_RETENTION=2592000000
 HEALTH_CHECK_INTERVAL=60000
 TIMEZONE_OFFSET=8  # UTC偏移小时数，默认+8（中国时区）
 METRICS_WINDOW=5  # 实时指标统计窗口（分钟），可选1-60，默认5分钟
+# 启动时清理残留的并发排队计数器（默认true，多实例部署时建议设为false）
+CLEAR_CONCURRENCY_QUEUES_ON_STARTUP=true
 
 # 🎨 Web 界面配置
 WEB_TITLE=Claude Relay Service

CLAUDE.md

@@ -22,6 +22,7 @@ Claude Relay Service 是一个多平台 AI API 中转服务，支持 **Claude (
 - **权限控制**: API Key支持权限配置（all/claude/gemini/openai等），控制可访问的服务类型
 - **客户端限制**: 基于User-Agent的客户端识别和限制，支持ClaudeCode、Gemini-CLI等预定义客户端
 - **模型黑名单**: 支持API Key级别的模型访问限制
+- **并发请求排队**: 当API Key并发数超限时，请求进入队列等待而非立即返回429，支持配置最大排队数、超时时间，适用于Claude Code Agent并行工具调用场景
 
 ### 主要服务组件
 
@@ -196,6 +197,7 @@ npm run service:stop          # 停止服务
 - `DEBUG_HTTP_TRAFFIC`: 启用HTTP请求/响应调试日志（默认false，仅开发环境）
 - `PROXY_USE_IPV4`: 代理使用IPv4（默认true）
 - `REQUEST_TIMEOUT`: 请求超时时间（毫秒，默认600000即10分钟）
+- `CLEAR_CONCURRENCY_QUEUES_ON_STARTUP`: 启动时清理残留的并发排队计数器（默认true，多实例部署时建议设为false）
 
 #### AWS Bedrock配置（可选）
 - `CLAUDE_CODE_USE_BEDROCK`: 启用Bedrock（设置为1启用）
@@ -343,6 +345,34 @@ npm run setup  # 自动生成密钥并创建管理员账户
 12. **成本统计不准确**: 运行 `npm run init:costs` 初始化成本数据，检查pricingService是否正确加载模型价格
 13. **缓存命中率低**: 查看缓存监控统计，调整LRU缓存大小配置
 14. **用户消息队列超时**: 优化后锁持有时间已从分钟级降到毫秒级（请求发送后立即释放），默认 `USER_MESSAGE_QUEUE_TIMEOUT_MS=5000` 已足够。如仍有超时，检查网络延迟或禁用此功能（`USER_MESSAGE_QUEUE_ENABLED=false`）
+15. **并发请求排队问题**:
+   - 排队超时：检查 `concurrentRequestQueueTimeoutMs` 配置是否合理（默认10秒）
+   - 排队数过多：调整 `concurrentRequestQueueMaxSize` 和 `concurrentRequestQueueMaxSizeMultiplier`
+   - 查看排队统计：访问 `/admin/concurrency-queue/stats` 接口查看 entered/success/timeout/cancelled/socket_changed/rejected_overload 统计
+   - 排队计数泄漏：系统重启时自动清理，或访问 `/admin/concurrency-queue` DELETE 接口手动清理
+   - Socket 身份验证失败：查看 `socket_changed` 统计，如果频繁发生，检查代理配置或客户端连接稳定性
+   - 健康检查拒绝：查看 `rejected_overload` 统计，表示队列过载时的快速失败次数
+
+### 代理配置要求（并发请求排队）
+
+使用并发请求排队功能时，需要正确配置代理（如 Nginx）的超时参数：
+
+- **推荐配置**: `proxy_read_timeout >= max(2 × concurrentRequestQueueTimeoutMs, 60s)`
+  - 当前默认排队超时 10 秒，Nginx 默认 `proxy_read_timeout = 60s` 已满足要求
+  - 如果调整排队超时到 60 秒，推荐代理超时 ≥ 120 秒
+- **Nginx 配置示例**:
+  ```nginx
+  location /api/ {
+      proxy_read_timeout 120s;  # 排队超时 60s 时推荐 120s
+      proxy_connect_timeout 10s;
+      # ...其他配置
+  }
+  ```
+- **企业防火墙环境**:
+  - 某些企业防火墙可能静默关闭长时间无数据的连接（20-40 秒）
+  - 如遇此问题，联系网络管理员调整空闲连接超时策略
+  - 或降低 `concurrentRequestQueueTimeoutMs` 配置
+- **后续升级说明**: 如有需要，后续版本可能提供可选的轻量级心跳机制
 
 ### 调试工具
 
@@ -455,6 +485,15 @@ npm run setup  # 自动生成密钥并创建管理员账户
 - **缓存优化**: 多层LRU缓存（解密缓存、账户缓存），全局缓存监控和统计
 - **成本追踪**: 实时token使用统计（input/output/cache_create/cache_read）和成本计算（基于pricingService）
 - **并发控制**: Redis Sorted Set实现的并发计数，支持自动过期清理
+- **并发请求排队**: 当API Key并发超限时，请求进入队列等待而非直接返回429
+  - **工作原理**: 采用「先占后检查」模式，每次轮询尝试占位，超限则释放继续等待
+  - **指数退避**: 初始200ms，指数增长至最大2秒，带±20%抖动防惊群效应
+  - **智能清理**: 排队计数有TTL保护（超时+30秒），进程崩溃也能自动清理
+  - **Socket身份验证**: 使用UUID token + socket对象引用双重验证，避免HTTP Keep-Alive连接复用导致的身份混淆
+  - **健康检查**: P90等待时间超过阈值时快速失败（返回429），避免新请求在过载时继续排队
+  - **配置参数**: `concurrentRequestQueueEnabled`（默认false）、`concurrentRequestQueueMaxSize`（默认3）、`concurrentRequestQueueMaxSizeMultiplier`（默认0）、`concurrentRequestQueueTimeoutMs`（默认10秒）、`concurrentRequestQueueMaxRedisFailCount`（默认5）、`concurrentRequestQueueHealthCheckEnabled`（默认true）、`concurrentRequestQueueHealthThreshold`（默认0.8）
+  - **最大排队数**: max(固定值, 并发限制×倍数)，例如并发限制=10、倍数=2时最大排队数=20
+  - **适用场景**: Claude Code Agent并行工具调用、批量请求处理
 - **客户端识别**: 基于User-Agent的客户端限制，支持预定义客户端（ClaudeCode、Gemini-CLI等）
 - **错误处理**: 529错误自动标记账户过载状态，配置时长内自动排除该账户
 
@@ -514,6 +553,11 @@ npm run setup  # 自动生成密钥并创建管理员账户
   - `overload:{accountId}` - 账户过载状态（529错误）
 - **并发控制**:
   - `concurrency:{accountId}` - Redis Sorted Set实现的并发计数
+- **并发请求排队**:
+  - `concurrency:queue:{apiKeyId}` - API Key级别的排队计数器（TTL由 `concurrentRequestQueueTimeoutMs` + 30秒缓冲决定）
+  - `concurrency:queue:stats:{apiKeyId}` - 排队统计（entered/success/timeout/cancelled）
+  - `concurrency:queue:wait_times:{apiKeyId}` - 按API Key的等待时间记录（用于P50/P90/P99计算）
+  - `concurrency:queue:wait_times:global` - 全局等待时间记录
 - **Webhook配置**:
   - `webhook_config:{id}` - Webhook配置
 - **用户消息队列**:

README.md

@@ -408,6 +408,8 @@ export ANTHROPIC_AUTH_TOKEN="后台创建的API密钥"
 
 如果该文件不存在，请手动创建。Windows 用户路径为 `C:\Users\你的用户名\.claude\config.json`。
 
+> 💡 **IntelliJ IDEA 用户推荐**：[Claude Code Plus](https://github.com/touwaeriol/claude-code-plus) - 将 Claude Code 直接集成到 IDE，支持代码理解、文件读写、命令执行。插件市场搜索 `Claude Code Plus` 即可安装。
+
 **Gemini CLI 设置环境变量：**
 
 **方式一（推荐）：通过 Gemini Assist API 方式访问**

VERSION

@@ -1 +1 @@
-1.1.232
+1.1.241

docker-compose.yml

@@ -21,6 +21,9 @@ services:
       - PORT=3000
       - HOST=0.0.0.0
       
+      # 🔧 请求体大小配置
+      - REQUEST_MAX_SIZE_MB=60
+      
       # 🔐 安全配置（必填）
       - JWT_SECRET=${JWT_SECRET}  # 必填：至少32字符的随机字符串
       - ENCRYPTION_KEY=${ENCRYPTION_KEY}  # 必填：32字符的加密密钥

package-lock.json

@@ -26,6 +26,7 @@
         "ioredis": "^5.3.2",
         "ldapjs": "^3.0.7",
         "morgan": "^1.10.0",
+        "node-cron": "^4.2.1",
         "nodemailer": "^7.0.6",
         "ora": "^5.4.1",
         "rate-limiter-flexible": "^5.0.5",
@@ -7034,6 +7035,15 @@
         "node": ">= 0.6"
       }
     },
+    "node_modules/node-cron": {
+      "version": "4.2.1",
+      "resolved": "https://registry.npmmirror.com/node-cron/-/node-cron-4.2.1.tgz",
+      "integrity": "sha512-lgimEHPE/QDgFlywTd8yTR61ptugX3Qer29efeyWw2rv259HtGBNn1vZVmp8lB9uo9wC0t/AT4iGqXxia+CJFg==",
+      "license": "ISC",
+      "engines": {
+        "node": ">=6.0.0"
+      }
+    },
     "node_modules/node-domexception": {
       "version": "1.0.0",
       "resolved": "https://registry.npmmirror.com/node-domexception/-/node-domexception-1.0.0.tgz",

package.json

@@ -65,6 +65,7 @@
     "ioredis": "^5.3.2",
     "ldapjs": "^3.0.7",
     "morgan": "^1.10.0",
+    "node-cron": "^4.2.1",
     "nodemailer": "^7.0.6",
     "ora": "^5.4.1",
     "rate-limiter-flexible": "^5.0.5",

pnpm-lock.yaml

@@ -59,6 +59,9 @@ importers:
       morgan:
         specifier: ^1.10.0
         version: 1.10.1
+      node-cron:
+        specifier: ^4.2.1
+        version: 4.2.1
       nodemailer:
         specifier: ^7.0.6
         version: 7.0.11
@@ -108,6 +111,9 @@ importers:
       prettier:
         specifier: ^3.6.2
         version: 3.7.4
+      prettier-plugin-tailwindcss:
+        specifier: ^0.7.2
+        version: 0.7.2(prettier@3.7.4)
       supertest:
         specifier: ^6.3.3
         version: 6.3.4
@@ -2144,6 +2150,10 @@ packages:
     resolution: {integrity: sha512-myRT3DiWPHqho5PrJaIRyaMv2kgYf0mUVgBNOYMuCH5Ki1yEiQaf/ZJuQ62nvpc44wL5WDbTX7yGJi1Neevw8w==}
     engines: {node: '>= 0.6'}
 
+  node-cron@4.2.1:
+    resolution: {integrity: sha512-lgimEHPE/QDgFlywTd8yTR61ptugX3Qer29efeyWw2rv259HtGBNn1vZVmp8lB9uo9wC0t/AT4iGqXxia+CJFg==}
+    engines: {node: '>=6.0.0'}
+
   node-domexception@1.0.0:
     resolution: {integrity: sha512-/jKZoMpw0F8GRwl4/eLROPA3cfcXtLApP0QzLmUT/HuPCZWyB7IY9ZrMeKw2O/nFIqPQB3PVM9aYm0F312AXDQ==}
     engines: {node: '>=10.5.0'}
@@ -2302,6 +2312,61 @@ packages:
     resolution: {integrity: sha512-GbK2cP9nraSSUF9N2XwUwqfzlAFlMNYYl+ShE/V+H8a9uNl/oUqB1w2EL54Jh0OlyRSd8RfWYJ3coVS4TROP2w==}
     engines: {node: '>=6.0.0'}
 
+  prettier-plugin-tailwindcss@0.7.2:
+    resolution: {integrity: sha512-LkphyK3Fw+q2HdMOoiEHWf93fNtYJwfamoKPl7UwtjFQdei/iIBoX11G6j706FzN3ymX9mPVi97qIY8328vdnA==}
+    engines: {node: '>=20.19'}
+    peerDependencies:
+      '@ianvs/prettier-plugin-sort-imports': '*'
+      '@prettier/plugin-hermes': '*'
+      '@prettier/plugin-oxc': '*'
+      '@prettier/plugin-pug': '*'
+      '@shopify/prettier-plugin-liquid': '*'
+      '@trivago/prettier-plugin-sort-imports': '*'
+      '@zackad/prettier-plugin-twig': '*'
+      prettier: ^3.0
+      prettier-plugin-astro: '*'
+      prettier-plugin-css-order: '*'
+      prettier-plugin-jsdoc: '*'
+      prettier-plugin-marko: '*'
+      prettier-plugin-multiline-arrays: '*'
+      prettier-plugin-organize-attributes: '*'
+      prettier-plugin-organize-imports: '*'
+      prettier-plugin-sort-imports: '*'
+      prettier-plugin-svelte: '*'
+    peerDependenciesMeta:
+      '@ianvs/prettier-plugin-sort-imports':
+        optional: true
+      '@prettier/plugin-hermes':
+        optional: true
+      '@prettier/plugin-oxc':
+        optional: true
+      '@prettier/plugin-pug':
+        optional: true
+      '@shopify/prettier-plugin-liquid':
+        optional: true
+      '@trivago/prettier-plugin-sort-imports':
+        optional: true
+      '@zackad/prettier-plugin-twig':
+        optional: true
+      prettier-plugin-astro:
+        optional: true
+      prettier-plugin-css-order:
+        optional: true
+      prettier-plugin-jsdoc:
+        optional: true
+      prettier-plugin-marko:
+        optional: true
+      prettier-plugin-multiline-arrays:
+        optional: true
+      prettier-plugin-organize-attributes:
+        optional: true
+      prettier-plugin-organize-imports:
+        optional: true
+      prettier-plugin-sort-imports:
+        optional: true
+      prettier-plugin-svelte:
+        optional: true
+
   prettier@3.7.4:
     resolution: {integrity: sha512-v6UNi1+3hSlVvv8fSaoUbggEM5VErKmmpGA7Pl3HF8V6uKY7rvClBOJlH6yNwQtfTueNkGVpOv/mtWL9L4bgRA==}
     engines: {node: '>=14'}
@@ -5692,6 +5757,8 @@ snapshots:
 
   negotiator@0.6.4: {}
 
+  node-cron@4.2.1: {}
+
   node-domexception@1.0.0: {}
 
   node-fetch@3.3.2:
@@ -5840,6 +5907,10 @@ snapshots:
     dependencies:
       fast-diff: 1.3.0
 
+  prettier-plugin-tailwindcss@0.7.2(prettier@3.7.4):
+    dependencies:
+      prettier: 3.7.4
+
   prettier@3.7.4: {}
 
   pretty-format@29.7.0:

src/app.js

@@ -68,6 +68,10 @@ class Application {
       logger.info('🔄 Initializing admin credentials...')
       await this.initializeAdmin()
 
+      // 🔒 安全启动：清理无效/伪造的管理员会话
+      logger.info('🔒 Cleaning up invalid admin sessions...')
+      await this.cleanupInvalidSessions()
+
       // 💰 初始化费用数据
       logger.info('💰 Checking cost data initialization...')
       const costInitService = require('./services/costInitService')
@@ -426,6 +430,54 @@ class Application {
     }
   }
 
+  // 🔒 清理无效/伪造的管理员会话（安全启动检查）
+  async cleanupInvalidSessions() {
+    try {
+      const client = redis.getClient()
+
+      // 获取所有 session:* 键
+      const sessionKeys = await client.keys('session:*')
+
+      let validCount = 0
+      let invalidCount = 0
+
+      for (const key of sessionKeys) {
+        // 跳过 admin_credentials（系统凭据）
+        if (key === 'session:admin_credentials') {
+          continue
+        }
+
+        const sessionData = await client.hgetall(key)
+
+        // 检查会话完整性：必须有 username 和 loginTime
+        const hasUsername = !!sessionData.username
+        const hasLoginTime = !!sessionData.loginTime
+
+        if (!hasUsername || !hasLoginTime) {
+          // 无效会话 - 可能是漏洞利用创建的伪造会话
+          invalidCount++
+          logger.security(
+            `🔒 Removing invalid session: ${key} (username: ${hasUsername}, loginTime: ${hasLoginTime})`
+          )
+          await client.del(key)
+        } else {
+          validCount++
+        }
+      }
+
+      if (invalidCount > 0) {
+        logger.security(`🔒 Startup security check: Removed ${invalidCount} invalid sessions`)
+      }
+
+      logger.success(
+        `✅ Session cleanup completed: ${validCount} valid, ${invalidCount} invalid removed`
+      )
+    } catch (error) {
+      // 清理失败不应阻止服务启动
+      logger.error('❌ Failed to cleanup invalid sessions:', error.message)
+    }
+  }
+
   // 🔍 Redis健康检查
   async checkRedisHealth() {
     try {
@@ -581,15 +633,40 @@ class Application {
 
         const now = Date.now()
         let totalCleaned = 0
+        let legacyCleaned = 0
 
         // 使用 Lua 脚本批量清理所有过期项
         for (const key of keys) {
+          // 跳过已知非 Sorted Set 类型的键（这些键有各自的清理逻辑）
+          // - concurrency:queue:stats:* 是 Hash 类型
+          // - concurrency:queue:wait_times:* 是 List 类型
+          // - concurrency:queue:* (不含stats/wait_times) 是 String 类型
+          if (
+            key.startsWith('concurrency:queue:stats:') ||
+            key.startsWith('concurrency:queue:wait_times:') ||
+            (key.startsWith('concurrency:queue:') &&
+              !key.includes(':stats:') &&
+              !key.includes(':wait_times:'))
+          ) {
+            continue
+          }
+
           try {
-            const cleaned = await redis.client.eval(
+            // 使用原子 Lua 脚本：先检查类型，再执行清理
+            // 返回值：0 = 正常清理无删除，1 = 清理后删除空键，-1 = 遗留键已删除
+            const result = await redis.client.eval(
               `
               local key = KEYS[1]
               local now = tonumber(ARGV[1])
 
+              -- 先检查键类型，只对 Sorted Set 执行清理
+              local keyType = redis.call('TYPE', key)
+              if keyType.ok ~= 'zset' then
+                -- 非 ZSET 类型的遗留键，直接删除
+                redis.call('DEL', key)
+                return -1
+              end
+
               -- 清理过期项
               redis.call('ZREMRANGEBYSCORE', key, '-inf', now)
 
@@ -608,8 +685,10 @@ class Application {
               key,
               now
             )
-            if (cleaned === 1) {
+            if (result === 1) {
               totalCleaned++
+            } else if (result === -1) {
+              legacyCleaned++
             }
           } catch (error) {
             logger.error(`❌ Failed to clean concurrency key ${key}:`, error)
@@ -619,6 +698,9 @@ class Application {
         if (totalCleaned > 0) {
           logger.info(`🔢 Concurrency cleanup: cleaned ${totalCleaned} expired keys`)
         }
+        if (legacyCleaned > 0) {
+          logger.warn(`🧹 Concurrency cleanup: removed ${legacyCleaned} legacy keys (wrong type)`)
+        }
       } catch (error) {
         logger.error('❌ Concurrency cleanup task failed:', error)
       }
@@ -633,6 +715,33 @@ class Application {
       // 然后启动定时清理任务
       userMessageQueueService.startCleanupTask()
     })
+
+    // 🚦 清理服务重启后残留的并发排队计数器
+    // 多实例部署时建议关闭此开关，避免新实例启动时清空其他实例的队列计数
+    // 可通过 DELETE /admin/concurrency/queue 接口手动清理
+    const clearQueuesOnStartup = process.env.CLEAR_CONCURRENCY_QUEUES_ON_STARTUP !== 'false'
+    if (clearQueuesOnStartup) {
+      redis.clearAllConcurrencyQueues().catch((error) => {
+        logger.error('❌ Error clearing concurrency queues on startup:', error)
+      })
+    } else {
+      logger.info(
+        '🚦 Skipping concurrency queue cleanup on startup (CLEAR_CONCURRENCY_QUEUES_ON_STARTUP=false)'
+      )
+    }
+
+    // 🧪 启动账户定时测试调度器
+    // 根据配置定期测试账户连通性并保存测试历史
+    const accountTestSchedulerEnabled =
+      process.env.ACCOUNT_TEST_SCHEDULER_ENABLED !== 'false' &&
+      config.accountTestScheduler?.enabled !== false
+    if (accountTestSchedulerEnabled) {
+      const accountTestSchedulerService = require('./services/accountTestSchedulerService')
+      accountTestSchedulerService.start()
+      logger.info('🧪 Account test scheduler service started')
+    } else {
+      logger.info('🧪 Account test scheduler service disabled')
+    }
   }
 
   setupGracefulShutdown() {
@@ -687,6 +796,15 @@ class Application {
             logger.error('❌ Error stopping cost rank service:', error)
           }
 
+          // 停止账户定时测试调度器
+          try {
+            const accountTestSchedulerService = require('./services/accountTestSchedulerService')
+            accountTestSchedulerService.stop()
+            logger.info('🧪 Account test scheduler service stopped')
+          } catch (error) {
+            logger.error('❌ Error stopping account test scheduler service:', error)
+          }
+
           // 🔢 清理所有并发计数（Phase 1 修复：防止重启泄漏）
           try {
             logger.info('🔢 Cleaning up all concurrency counters...')

src/middleware/auth.js

@@ -8,6 +8,102 @@ const redis = require('../models/redis')
 const ClientValidator = require('../validators/clientValidator')
 const ClaudeCodeValidator = require('../validators/clients/claudeCodeValidator')
 const claudeRelayConfigService = require('../services/claudeRelayConfigService')
+const { calculateWaitTimeStats } = require('../utils/statsHelper')
+
+// 工具函数
+function sleep(ms) {
+  return new Promise((resolve) => setTimeout(resolve, ms))
+}
+
+/**
+ * 检查排队是否过载，决定是否应该快速失败
+ * 详见 design.md Decision 7: 排队健康检查与快速失败
+ *
+ * @param {string} apiKeyId - API Key ID
+ * @param {number} timeoutMs - 排队超时时间（毫秒）
+ * @param {Object} queueConfig - 队列配置
+ * @param {number} maxQueueSize - 最大排队数
+ * @returns {Promise<Object>} { reject: boolean, reason?: string, estimatedWaitMs?: number, timeoutMs?: number }
+ */
+async function shouldRejectDueToOverload(apiKeyId, timeoutMs, queueConfig, maxQueueSize) {
+  try {
+    // 如果健康检查被禁用，直接返回不拒绝
+    if (!queueConfig.concurrentRequestQueueHealthCheckEnabled) {
+      return { reject: false, reason: 'health_check_disabled' }
+    }
+
+    // 🔑 先检查当前队列长度
+    const currentQueueCount = await redis.getConcurrencyQueueCount(apiKeyId).catch(() => 0)
+
+    // 队列为空，说明系统已恢复，跳过健康检查
+    if (currentQueueCount === 0) {
+      return { reject: false, reason: 'queue_empty', currentQueueCount: 0 }
+    }
+
+    // 🔑 关键改进：只有当队列接近满载时才进行健康检查
+    // 队列长度 <= maxQueueSize * 0.5 时，认为系统有足够余量，跳过健康检查
+    // 这避免了在队列较短时过于保守地拒绝请求
+    // 使用 ceil 确保小队列（如 maxQueueSize=3）时阈值为 2，即队列 <=1 时跳过
+    const queueLoadThreshold = Math.ceil(maxQueueSize * 0.5)
+    if (currentQueueCount <= queueLoadThreshold) {
+      return {
+        reject: false,
+        reason: 'queue_not_loaded',
+        currentQueueCount,
+        queueLoadThreshold,
+        maxQueueSize
+      }
+    }
+
+    // 获取该 API Key 的等待时间样本
+    const waitTimes = await redis.getQueueWaitTimes(apiKeyId)
+    const stats = calculateWaitTimeStats(waitTimes)
+
+    // 样本不足（< 10），跳过健康检查，避免冷启动误判
+    if (!stats || stats.sampleCount < 10) {
+      return { reject: false, reason: 'insufficient_samples', sampleCount: stats?.sampleCount || 0 }
+    }
+
+    // P90 不可靠时也跳过（虽然 sampleCount >= 10 时 p90Unreliable 应该是 false）
+    if (stats.p90Unreliable) {
+      return { reject: false, reason: 'p90_unreliable', sampleCount: stats.sampleCount }
+    }
+
+    // 计算健康阈值：P90 >= 超时时间 × 阈值 时拒绝
+    const threshold = queueConfig.concurrentRequestQueueHealthThreshold || 0.8
+    const maxAllowedP90 = timeoutMs * threshold
+
+    if (stats.p90 >= maxAllowedP90) {
+      return {
+        reject: true,
+        reason: 'queue_overloaded',
+        estimatedWaitMs: stats.p90,
+        timeoutMs,
+        threshold,
+        sampleCount: stats.sampleCount,
+        currentQueueCount,
+        maxQueueSize
+      }
+    }
+
+    return { reject: false, p90: stats.p90, sampleCount: stats.sampleCount, currentQueueCount }
+  } catch (error) {
+    // 健康检查出错时不阻塞请求，记录警告并继续
+    logger.warn(`Health check failed for ${apiKeyId}:`, error.message)
+    return { reject: false, reason: 'health_check_error', error: error.message }
+  }
+}
+
+// 排队轮询配置常量（可通过配置文件覆盖）
+// 性能权衡：初始间隔越短响应越快，但 Redis QPS 越高
+// 当前配置：100 个等待者时约 250-300 QPS（指数退避后）
+const QUEUE_POLLING_CONFIG = {
+  pollIntervalMs: 200, // 初始轮询间隔（毫秒）- 平衡响应速度和 Redis 压力
+  maxPollIntervalMs: 2000, // 最大轮询间隔（毫秒）- 长时间等待时降低 Redis 压力
+  backoffFactor: 1.5, // 指数退避系数
+  jitterRatio: 0.2, // 抖动比例（±20%）- 防止惊群效应
+  maxRedisFailCount: 5 // 连续 Redis 失败阈值（从 3 提高到 5，提高网络抖动容忍度）
+}
 
 const FALLBACK_CONCURRENCY_CONFIG = {
   leaseSeconds: 300,
@@ -128,9 +224,223 @@ function isTokenCountRequest(req) {
   return false
 }
 
+/**
+ * 等待并发槽位（排队机制核心）
+ *
+ * 采用「先占后检查」模式避免竞态条件：
+ * - 每次轮询时尝试 incrConcurrency 占位
+ * - 如果超限则 decrConcurrency 释放并继续等待
+ * - 成功获取槽位后返回，调用方无需再次 incrConcurrency
+ *
+ * ⚠️ 重要清理责任说明：
+ * - 排队计数：此函数的 finally 块负责调用 decrConcurrencyQueue 清理
+ * - 并发槽位：当返回 acquired=true 时，槽位已被占用（通过 incrConcurrency）
+ *   调用方必须在请求结束时调用 decrConcurrency 释放槽位
+ *   （已在 authenticateApiKey 的 finally 块中处理）
+ *
+ * @param {Object} req - Express 请求对象
+ * @param {Object} res - Express 响应对象
+ * @param {string} apiKeyId - API Key ID
+ * @param {Object} queueOptions - 配置参数
+ * @returns {Promise<Object>} { acquired: boolean, reason?: string, waitTimeMs: number }
+ */
+async function waitForConcurrencySlot(req, res, apiKeyId, queueOptions) {
+  const {
+    concurrencyLimit,
+    requestId,
+    leaseSeconds,
+    timeoutMs,
+    pollIntervalMs,
+    maxPollIntervalMs,
+    backoffFactor,
+    jitterRatio,
+    maxRedisFailCount: configMaxRedisFailCount
+  } = queueOptions
+
+  let clientDisconnected = false
+  // 追踪轮询过程中是否临时占用了槽位（用于异常时清理）
+  // 工作流程：
+  // 1. incrConcurrency 成功且 count <= limit 时，设置 internalSlotAcquired = true
+  // 2. 统计记录完成后，设置 internalSlotAcquired = false 并返回（所有权转移给调用方）
+  // 3. 如果在步骤 1-2 之间发生异常，finally 块会检测到 internalSlotAcquired = true 并释放槽位
+  let internalSlotAcquired = false
+
+  // 监听客户端断开事件
+  // ⚠️ 重要：必须监听 socket 的事件，而不是 req 的事件！
+  // 原因：对于 POST 请求，当 body-parser 读取完请求体后，req（IncomingMessage 可读流）
+  // 的 'close' 事件会立即触发，但这不代表客户端断开连接！客户端仍在等待响应。
+  // socket 的 'close' 事件才是真正的连接关闭信号。
+  const { socket } = req
+  const onSocketClose = () => {
+    clientDisconnected = true
+    logger.debug(
+      `🔌 [Queue] Socket closed during queue wait for API key ${apiKeyId}, requestId: ${requestId}`
+    )
+  }
+
+  if (socket) {
+    socket.once('close', onSocketClose)
+  }
+
+  // 检查 socket 是否在监听器注册前已被销毁（边界情况）
+  if (socket?.destroyed) {
+    clientDisconnected = true
+  }
+
+  const startTime = Date.now()
+  let pollInterval = pollIntervalMs
+  let redisFailCount = 0
+  // 优先使用配置中的值，否则使用默认值
+  const maxRedisFailCount = configMaxRedisFailCount || QUEUE_POLLING_CONFIG.maxRedisFailCount
+
+  try {
+    while (Date.now() - startTime < timeoutMs) {
+      // 检测客户端是否断开（双重检查：事件标记 + socket 状态）
+      // socket.destroyed 是同步检查，确保即使事件处理有延迟也能及时检测
+      if (clientDisconnected || socket?.destroyed) {
+        redis
+          .incrConcurrencyQueueStats(apiKeyId, 'cancelled')
+          .catch((e) => logger.warn('Failed to record cancelled stat:', e))
+        return {
+          acquired: false,
+          reason: 'client_disconnected',
+          waitTimeMs: Date.now() - startTime
+        }
+      }
+
+      // 尝试获取槽位（先占后检查）
+      try {
+        const count = await redis.incrConcurrency(apiKeyId, requestId, leaseSeconds)
+        redisFailCount = 0 // 重置失败计数
+
+        if (count <= concurrencyLimit) {
+          // 成功获取槽位！
+          const waitTimeMs = Date.now() - startTime
+
+          // 槽位所有权转移说明：
+          // 1. 此时槽位已通过 incrConcurrency 获取
+          // 2. 先标记 internalSlotAcquired = true，确保异常时 finally 块能清理
+          // 3. 统计操作完成后，清除标记并返回，所有权转移给调用方
+          // 4. 调用方（authenticateApiKey）负责在请求结束时释放槽位
+
+          // 标记槽位已获取（用于异常时 finally 块清理）
+          internalSlotAcquired = true
+
+          // 记录统计（非阻塞，fire-and-forget 模式）
+          // ⚠️ 设计说明：
+          // - 故意不 await 这些 Promise，因为统计记录不应阻塞请求处理
+          // - 每个 Promise 都有独立的 .catch()，确保单个失败不影响其他
+          // - 外层 .catch() 是防御性措施，处理 Promise.all 本身的异常
+          // - 即使统计记录在函数返回后才完成/失败，也是安全的（仅日志记录）
+          // - 统计数据丢失可接受，不影响核心业务逻辑
+          Promise.all([
+            redis
+              .recordQueueWaitTime(apiKeyId, waitTimeMs)
+              .catch((e) => logger.warn('Failed to record queue wait time:', e)),
+            redis
+              .recordGlobalQueueWaitTime(waitTimeMs)
+              .catch((e) => logger.warn('Failed to record global wait time:', e)),
+            redis
+              .incrConcurrencyQueueStats(apiKeyId, 'success')
+              .catch((e) => logger.warn('Failed to increment success stats:', e))
+          ]).catch((e) => logger.warn('Failed to record queue stats batch:', e))
+
+          // 成功返回前清除标记（所有权转移给调用方，由其负责释放）
+          internalSlotAcquired = false
+          return { acquired: true, waitTimeMs }
+        }
+
+        // 超限，释放槽位继续等待
+        try {
+          await redis.decrConcurrency(apiKeyId, requestId)
+        } catch (decrError) {
+          // 释放失败时记录警告但继续轮询
+          // 下次 incrConcurrency 会自然覆盖同一 requestId 的条目
+          logger.warn(
+            `Failed to release slot during polling for ${apiKeyId}, will retry:`,
+            decrError
+          )
+        }
+      } catch (redisError) {
+        redisFailCount++
+        logger.error(
+          `Redis error in queue polling (${redisFailCount}/${maxRedisFailCount}):`,
+          redisError
+        )
+
+        if (redisFailCount >= maxRedisFailCount) {
+          // 连续 Redis 失败，放弃排队
+          return {
+            acquired: false,
+            reason: 'redis_error',
+            waitTimeMs: Date.now() - startTime
+          }
+        }
+      }
+
+      // 指数退避等待
+      await sleep(pollInterval)
+
+      // 计算下一次轮询间隔（指数退避 + 抖动）
+      // 1. 先应用指数退避
+      let nextInterval = pollInterval * backoffFactor
+      // 2. 添加抖动防止惊群效应（±jitterRatio 范围内的随机偏移）
+      //    抖动范围：[-jitterRatio, +jitterRatio]，例如 jitterRatio=0.2 时为 ±20%
+      //    这是预期行为：负抖动可使间隔略微缩短，正抖动可使间隔略微延长
+      //    目的是分散多个等待者的轮询时间点，避免同时请求 Redis
+      const jitter = nextInterval * jitterRatio * (Math.random() * 2 - 1)
+      nextInterval = nextInterval + jitter
+      // 3. 确保在合理范围内：最小 1ms，最大 maxPollIntervalMs
+      //    Math.max(1, ...) 保证即使负抖动也不会产生 ≤0 的间隔
+      pollInterval = Math.max(1, Math.min(nextInterval, maxPollIntervalMs))
+    }
+
+    // 超时
+    redis
+      .incrConcurrencyQueueStats(apiKeyId, 'timeout')
+      .catch((e) => logger.warn('Failed to record timeout stat:', e))
+    return { acquired: false, reason: 'timeout', waitTimeMs: Date.now() - startTime }
+  } finally {
+    // 确保清理：
+    // 1. 减少排队计数（排队计数在调用方已增加，这里负责减少）
+    try {
+      await redis.decrConcurrencyQueue(apiKeyId)
+    } catch (cleanupError) {
+      // 清理失败记录错误（可能导致计数泄漏，但有 TTL 保护）
+      logger.error(
+        `Failed to decrement queue count in finally block for ${apiKeyId}:`,
+        cleanupError
+      )
+    }
+
+    // 2. 如果内部获取了槽位但未正常返回（异常路径），释放槽位
+    if (internalSlotAcquired) {
+      try {
+        await redis.decrConcurrency(apiKeyId, requestId)
+        logger.warn(
+          `⚠️ Released orphaned concurrency slot in finally block for ${apiKeyId}, requestId: ${requestId}`
+        )
+      } catch (slotCleanupError) {
+        logger.error(
+          `Failed to release orphaned concurrency slot for ${apiKeyId}:`,
+          slotCleanupError
+        )
+      }
+    }
+
+    // 清理 socket 事件监听器
+    if (socket) {
+      socket.removeListener('close', onSocketClose)
+    }
+  }
+}
+
 // 🔑 API Key验证中间件（优化版）
 const authenticateApiKey = async (req, res, next) => {
   const startTime = Date.now()
+  let authErrored = false
+  let concurrencyCleanup = null
+  let hasConcurrencySlot = false
 
   try {
     // 安全提取API Key，支持多种格式（包括Gemini CLI支持）
@@ -265,39 +575,346 @@ const authenticateApiKey = async (req, res, next) => {
       }
       const requestId = uuidv4()
 
+      // ⚠️ 优化后的 Connection: close 设置策略
+      // 问题背景：HTTP Keep-Alive 使多个请求共用同一个 TCP 连接
+      // 当第一个请求正在处理，第二个请求进入排队时，它们共用同一个 socket
+      // 如果客户端超时关闭连接，两个请求都会受影响
+      // 优化方案：只有在请求实际进入排队时才设置 Connection: close
+      // 未排队的请求保持 Keep-Alive，避免不必要的 TCP 握手开销
+      // 详见 design.md Decision 2: Connection: close 设置时机
+      // 注意：Connection: close 将在下方代码实际进入排队时设置（第 637 行左右）
+
+      // ============================================================
+      // 🔒 并发槽位状态管理说明
+      // ============================================================
+      // 此函数中有两个关键状态变量：
+      // - hasConcurrencySlot: 当前是否持有并发槽位
+      // - concurrencyCleanup: 错误时调用的清理函数
+      //
+      // 状态转换流程：
+      // 1. incrConcurrency 成功 → hasConcurrencySlot=true, 设置临时清理函数
+      // 2. 若超限 → 释放槽位，hasConcurrencySlot=false, concurrencyCleanup=null
+      // 3. 若排队成功 → hasConcurrencySlot=true, 升级为完整清理函数（含 interval 清理）
+      // 4. 请求结束（res.close/req.close）→ 调用 decrementConcurrency 释放
+      // 5. 认证错误 → finally 块调用 concurrencyCleanup 释放
+      //
+      // 为什么需要两种清理函数？
+      // - 临时清理：在排队/认证过程中出错时使用，只释放槽位
+      // - 完整清理：请求正常开始后使用，还需清理 leaseRenewInterval
+      // ============================================================
+      const setTemporaryConcurrencyCleanup = () => {
+        concurrencyCleanup = async () => {
+          if (!hasConcurrencySlot) {
+            return
+          }
+          hasConcurrencySlot = false
+          try {
+            await redis.decrConcurrency(validation.keyData.id, requestId)
+          } catch (cleanupError) {
+            logger.error(
+              `Failed to decrement concurrency after auth error for key ${validation.keyData.id}:`,
+              cleanupError
+            )
+          }
+        }
+      }
+
       const currentConcurrency = await redis.incrConcurrency(
         validation.keyData.id,
         requestId,
         leaseSeconds
       )
+      hasConcurrencySlot = true
+      setTemporaryConcurrencyCleanup()
       logger.api(
         `📈 Incremented concurrency for key: ${validation.keyData.id} (${validation.keyData.name}), current: ${currentConcurrency}, limit: ${concurrencyLimit}`
       )
 
       if (currentConcurrency > concurrencyLimit) {
-        // 如果超过限制，立即减少计数（添加 try-catch 防止异常导致并发泄漏）
+        // 1. 先释放刚占用的槽位
         try {
-          const newCount = await redis.decrConcurrency(validation.keyData.id, requestId)
-          logger.api(
-            `📉 Decremented concurrency (429 rejected) for key: ${validation.keyData.id} (${validation.keyData.name}), new count: ${newCount}`
-          )
+          await redis.decrConcurrency(validation.keyData.id, requestId)
         } catch (error) {
           logger.error(
             `Failed to decrement concurrency after limit exceeded for key ${validation.keyData.id}:`,
             error
           )
         }
-        logger.security(
-          `🚦 Concurrency limit exceeded for key: ${validation.keyData.id} (${
-            validation.keyData.name
-          }), current: ${currentConcurrency - 1}, limit: ${concurrencyLimit}`
+        hasConcurrencySlot = false
+        concurrencyCleanup = null
+
+        // 2. 获取排队配置
+        const queueConfig = await claudeRelayConfigService.getConfig()
+
+        // 3. 排队功能未启用，直接返回 429（保持现有行为）
+        if (!queueConfig.concurrentRequestQueueEnabled) {
+          logger.security(
+            `🚦 Concurrency limit exceeded for key: ${validation.keyData.id} (${
+              validation.keyData.name
+            }), current: ${currentConcurrency - 1}, limit: ${concurrencyLimit}`
+          )
+          // 建议客户端在短暂延迟后重试（并发场景下通常很快会有槽位释放）
+          res.set('Retry-After', '1')
+          return res.status(429).json({
+            error: 'Concurrency limit exceeded',
+            message: `Too many concurrent requests. Limit: ${concurrencyLimit} concurrent requests`,
+            currentConcurrency: currentConcurrency - 1,
+            concurrencyLimit
+          })
+        }
+
+        // 4. 计算最大排队数
+        const maxQueueSize = Math.max(
+          concurrencyLimit * queueConfig.concurrentRequestQueueMaxSizeMultiplier,
+          queueConfig.concurrentRequestQueueMaxSize
         )
-        return res.status(429).json({
-          error: 'Concurrency limit exceeded',
-          message: `Too many concurrent requests. Limit: ${concurrencyLimit} concurrent requests`,
-          currentConcurrency: currentConcurrency - 1,
-          concurrencyLimit
-        })
+
+        // 4.5 排队健康检查：过载时快速失败
+        // 详见 design.md Decision 7: 排队健康检查与快速失败
+        const overloadCheck = await shouldRejectDueToOverload(
+          validation.keyData.id,
+          queueConfig.concurrentRequestQueueTimeoutMs,
+          queueConfig,
+          maxQueueSize
+        )
+        if (overloadCheck.reject) {
+          // 使用健康检查返回的当前排队数，避免重复调用 Redis
+          const currentQueueCount = overloadCheck.currentQueueCount || 0
+          logger.api(
+            `🚨 Queue overloaded for key: ${validation.keyData.id} (${validation.keyData.name}), ` +
+              `P90=${overloadCheck.estimatedWaitMs}ms, timeout=${overloadCheck.timeoutMs}ms, ` +
+              `threshold=${overloadCheck.threshold}, samples=${overloadCheck.sampleCount}, ` +
+              `concurrency=${concurrencyLimit}, queue=${currentQueueCount}/${maxQueueSize}`
+          )
+          // 记录被拒绝的过载统计
+          redis
+            .incrConcurrencyQueueStats(validation.keyData.id, 'rejected_overload')
+            .catch((e) => logger.warn('Failed to record rejected_overload stat:', e))
+          // 返回 429 + Retry-After，让客户端稍后重试
+          const retryAfterSeconds = 30
+          res.set('Retry-After', String(retryAfterSeconds))
+          return res.status(429).json({
+            error: 'Queue overloaded',
+            message: `Queue is overloaded. Estimated wait time (${overloadCheck.estimatedWaitMs}ms) exceeds threshold. Limit: ${concurrencyLimit} concurrent requests, queue: ${currentQueueCount}/${maxQueueSize}. Please retry later.`,
+            currentConcurrency: concurrencyLimit,
+            concurrencyLimit,
+            queueCount: currentQueueCount,
+            maxQueueSize,
+            estimatedWaitMs: overloadCheck.estimatedWaitMs,
+            timeoutMs: overloadCheck.timeoutMs,
+            queueTimeoutMs: queueConfig.concurrentRequestQueueTimeoutMs,
+            retryAfterSeconds
+          })
+        }
+
+        // 5. 尝试进入排队（原子操作：先增加再检查，避免竞态条件）
+        let queueIncremented = false
+        try {
+          const newQueueCount = await redis.incrConcurrencyQueue(
+            validation.keyData.id,
+            queueConfig.concurrentRequestQueueTimeoutMs
+          )
+          queueIncremented = true
+
+          if (newQueueCount > maxQueueSize) {
+            // 超过最大排队数，立即释放并返回 429
+            await redis.decrConcurrencyQueue(validation.keyData.id)
+            queueIncremented = false
+            logger.api(
+              `🚦 Concurrency queue full for key: ${validation.keyData.id} (${validation.keyData.name}), ` +
+                `queue: ${newQueueCount - 1}, maxQueue: ${maxQueueSize}`
+            )
+            // 队列已满，建议客户端在排队超时时间后重试
+            const retryAfterSeconds = Math.ceil(queueConfig.concurrentRequestQueueTimeoutMs / 1000)
+            res.set('Retry-After', String(retryAfterSeconds))
+            return res.status(429).json({
+              error: 'Concurrency queue full',
+              message: `Too many requests waiting in queue. Limit: ${concurrencyLimit} concurrent requests, queue: ${newQueueCount - 1}/${maxQueueSize}, timeout: ${retryAfterSeconds}s`,
+              currentConcurrency: concurrencyLimit,
+              concurrencyLimit,
+              queueCount: newQueueCount - 1,
+              maxQueueSize,
+              queueTimeoutMs: queueConfig.concurrentRequestQueueTimeoutMs,
+              retryAfterSeconds
+            })
+          }
+
+          // 6. 已成功进入排队，记录统计并开始等待槽位
+          logger.api(
+            `⏳ Request entering queue for key: ${validation.keyData.id} (${validation.keyData.name}), ` +
+              `queue position: ${newQueueCount}`
+          )
+          redis
+            .incrConcurrencyQueueStats(validation.keyData.id, 'entered')
+            .catch((e) => logger.warn('Failed to record entered stat:', e))
+
+          // ⚠️ 仅在请求实际进入排队时设置 Connection: close
+          // 详见 design.md Decision 2: Connection: close 设置时机
+          // 未排队的请求保持 Keep-Alive，避免不必要的 TCP 握手开销
+          if (!res.headersSent) {
+            res.setHeader('Connection', 'close')
+            logger.api(
+              `🔌 [Queue] Set Connection: close for queued request, key: ${validation.keyData.id}`
+            )
+          }
+
+          // ⚠️ 记录排队开始时的 socket 标识，用于排队完成后验证
+          // 问题背景：HTTP Keep-Alive 连接复用时，长时间排队可能导致 socket 被其他请求使用
+          // 验证方法：使用 UUID token + socket 对象引用双重验证
+          // 详见 design.md Decision 1: Socket 身份验证机制
+          req._crService = req._crService || {}
+          req._crService.queueToken = uuidv4()
+          req._crService.originalSocket = req.socket
+          req._crService.startTime = Date.now()
+          const savedToken = req._crService.queueToken
+          const savedSocket = req._crService.originalSocket
+
+          // ⚠️ 重要：在调用前将 queueIncremented 设为 false
+          // 因为 waitForConcurrencySlot 的 finally 块会负责清理排队计数
+          // 如果在调用后设置，当 waitForConcurrencySlot 抛出异常时
+          // 外层 catch 块会重复减少计数（finally 已经减过一次）
+          queueIncremented = false
+
+          const slot = await waitForConcurrencySlot(req, res, validation.keyData.id, {
+            concurrencyLimit,
+            requestId,
+            leaseSeconds,
+            timeoutMs: queueConfig.concurrentRequestQueueTimeoutMs,
+            pollIntervalMs: QUEUE_POLLING_CONFIG.pollIntervalMs,
+            maxPollIntervalMs: QUEUE_POLLING_CONFIG.maxPollIntervalMs,
+            backoffFactor: QUEUE_POLLING_CONFIG.backoffFactor,
+            jitterRatio: QUEUE_POLLING_CONFIG.jitterRatio,
+            maxRedisFailCount: queueConfig.concurrentRequestQueueMaxRedisFailCount
+          })
+
+          // 7. 处理排队结果
+          if (!slot.acquired) {
+            if (slot.reason === 'client_disconnected') {
+              // 客户端已断开，不返回响应（连接已关闭）
+              logger.api(
+                `🔌 Client disconnected while queuing for key: ${validation.keyData.id} (${validation.keyData.name})`
+              )
+              return
+            }
+
+            if (slot.reason === 'redis_error') {
+              // Redis 连续失败，返回 503
+              logger.error(
+                `❌ Redis error during queue wait for key: ${validation.keyData.id} (${validation.keyData.name})`
+              )
+              return res.status(503).json({
+                error: 'Service temporarily unavailable',
+                message: 'Failed to acquire concurrency slot due to internal error'
+              })
+            }
+            // 排队超时（使用 api 级别，与其他排队日志保持一致）
+            logger.api(
+              `⏰ Queue timeout for key: ${validation.keyData.id} (${validation.keyData.name}), waited: ${slot.waitTimeMs}ms`
+            )
+            // 已等待超时，建议客户端稍后重试
+            // ⚠️ Retry-After 策略优化：
+            // - 请求已经等了完整的 timeout 时间，说明系统负载较高
+            // - 过早重试（如固定 5 秒）会加剧拥塞，导致更多超时
+            // - 合理策略：使用 timeout 时间的一半作为重试间隔
+            // - 最小值 5 秒，最大值 30 秒，避免极端情况
+            const timeoutSeconds = Math.ceil(queueConfig.concurrentRequestQueueTimeoutMs / 1000)
+            const retryAfterSeconds = Math.max(5, Math.min(30, Math.ceil(timeoutSeconds / 2)))
+            res.set('Retry-After', String(retryAfterSeconds))
+            return res.status(429).json({
+              error: 'Queue timeout',
+              message: `Request timed out waiting for concurrency slot. Limit: ${concurrencyLimit} concurrent requests, maxQueue: ${maxQueueSize}, Queue timeout: ${timeoutSeconds}s, waited: ${slot.waitTimeMs}ms`,
+              currentConcurrency: concurrencyLimit,
+              concurrencyLimit,
+              maxQueueSize,
+              queueTimeoutMs: queueConfig.concurrentRequestQueueTimeoutMs,
+              waitTimeMs: slot.waitTimeMs,
+              retryAfterSeconds
+            })
+          }
+
+          // 8. 排队成功，slot.acquired 表示已在 waitForConcurrencySlot 中获取到槽位
+          logger.api(
+            `✅ Queue wait completed for key: ${validation.keyData.id} (${validation.keyData.name}), ` +
+              `waited: ${slot.waitTimeMs}ms`
+          )
+          hasConcurrencySlot = true
+          setTemporaryConcurrencyCleanup()
+
+          // 9. ⚠️ 关键检查：排队等待结束后，验证客户端是否还在等待响应
+          // 长时间排队后，客户端可能在应用层已放弃（如 Claude Code 的超时机制），
+          // 但 TCP 连接仍然存活。此时继续处理请求是浪费资源。
+          // 注意：如果发送了心跳，headersSent 会是 true，但这是正常的
+          const postQueueSocket = req.socket
+          // 只检查连接是否真正断开（destroyed/writableEnded/socketDestroyed）
+          // headersSent 在心跳场景下是正常的，不应该作为放弃的依据
+          if (res.destroyed || res.writableEnded || postQueueSocket?.destroyed) {
+            logger.warn(
+              `⚠️ Client no longer waiting after queue for key: ${validation.keyData.id} (${validation.keyData.name}), ` +
+                `waited: ${slot.waitTimeMs}ms | destroyed: ${res.destroyed}, ` +
+                `writableEnded: ${res.writableEnded}, socketDestroyed: ${postQueueSocket?.destroyed}`
+            )
+            // 释放刚获取的槽位
+            hasConcurrencySlot = false
+            await redis
+              .decrConcurrency(validation.keyData.id, requestId)
+              .catch((e) => logger.error('Failed to release slot after client abandoned:', e))
+            // 不返回响应（客户端已不在等待）
+            return
+          }
+
+          // 10. ⚠️ 关键检查：验证 socket 身份是否改变
+          // HTTP Keep-Alive 连接复用可能导致排队期间 socket 被其他请求使用
+          // 验证方法：UUID token + socket 对象引用双重验证
+          // 详见 design.md Decision 1: Socket 身份验证机制
+          const queueData = req._crService
+          const socketIdentityChanged =
+            !queueData ||
+            queueData.queueToken !== savedToken ||
+            queueData.originalSocket !== savedSocket
+
+          if (socketIdentityChanged) {
+            logger.error(
+              `❌ [Queue] Socket identity changed during queue wait! ` +
+                `key: ${validation.keyData.id} (${validation.keyData.name}), ` +
+                `waited: ${slot.waitTimeMs}ms | ` +
+                `tokenMatch: ${queueData?.queueToken === savedToken}, ` +
+                `socketMatch: ${queueData?.originalSocket === savedSocket}`
+            )
+            // 释放刚获取的槽位
+            hasConcurrencySlot = false
+            await redis
+              .decrConcurrency(validation.keyData.id, requestId)
+              .catch((e) => logger.error('Failed to release slot after socket identity change:', e))
+            // 记录 socket_changed 统计
+            redis
+              .incrConcurrencyQueueStats(validation.keyData.id, 'socket_changed')
+              .catch((e) => logger.warn('Failed to record socket_changed stat:', e))
+            // 不返回响应（socket 已被其他请求使用）
+            return
+          }
+        } catch (queueError) {
+          // 异常时清理资源，防止泄漏
+          // 1. 清理排队计数（如果还没被 waitForConcurrencySlot 的 finally 清理）
+          if (queueIncremented) {
+            await redis
+              .decrConcurrencyQueue(validation.keyData.id)
+              .catch((e) => logger.error('Failed to cleanup queue count after error:', e))
+          }
+
+          // 2. 防御性清理：如果 waitForConcurrencySlot 内部获取了槽位但在返回前异常
+          //    虽然这种情况极少发生（统计记录的异常会被内部捕获），但为了安全起见
+          //    尝试释放可能已获取的槽位。decrConcurrency 使用 ZREM，即使成员不存在也安全
+          if (hasConcurrencySlot) {
+            hasConcurrencySlot = false
+            await redis
+              .decrConcurrency(validation.keyData.id, requestId)
+              .catch((e) =>
+                logger.error('Failed to cleanup concurrency slot after queue error:', e)
+              )
+          }
+
+          throw queueError
+        }
       }
 
       const renewIntervalMs =
@@ -358,6 +975,7 @@ const authenticateApiKey = async (req, res, next) => {
       const decrementConcurrency = async () => {
         if (!concurrencyDecremented) {
           concurrencyDecremented = true
+          hasConcurrencySlot = false
           if (leaseRenewInterval) {
             clearInterval(leaseRenewInterval)
             leaseRenewInterval = null
@@ -372,6 +990,11 @@ const authenticateApiKey = async (req, res, next) => {
           }
         }
       }
+      // 升级为完整清理函数（包含 leaseRenewInterval 清理逻辑）
+      // 此时请求已通过认证，后续由 res.close/req.close 事件触发清理
+      if (hasConcurrencySlot) {
+        concurrencyCleanup = decrementConcurrency
+      }
 
       // 监听最可靠的事件（避免重复监听）
       // res.on('close') 是最可靠的，会在连接关闭时触发
@@ -697,6 +1320,7 @@ const authenticateApiKey = async (req, res, next) => {
 
     return next()
   } catch (error) {
+    authErrored = true
     const authDuration = Date.now() - startTime
     logger.error(`❌ Authentication middleware error (${authDuration}ms):`, {
       error: error.message,
@@ -710,6 +1334,14 @@ const authenticateApiKey = async (req, res, next) => {
       error: 'Authentication error',
       message: 'Internal server error during authentication'
     })
+  } finally {
+    if (authErrored && typeof concurrencyCleanup === 'function') {
+      try {
+        await concurrencyCleanup()
+      } catch (cleanupError) {
+        logger.error('Failed to cleanup concurrency after auth error:', cleanupError)
+      }
+    }
   }
 }
 
@@ -757,6 +1389,18 @@ const authenticateAdmin = async (req, res, next) => {
       })
     }
 
+    // 🔒 安全修复：验证会话必须字段（防止伪造会话绕过认证）
+    if (!adminSession.username || !adminSession.loginTime) {
+      logger.security(
+        `🔒 Corrupted admin session from ${req.ip || 'unknown'} - missing required fields (username: ${!!adminSession.username}, loginTime: ${!!adminSession.loginTime})`
+      )
+      await redis.deleteSession(token) // 清理无效/伪造的会话
+      return res.status(401).json({
+        error: 'Invalid session',
+        message: 'Session data corrupted or incomplete'
+      })
+    }
+
     // 检查会话活跃性（可选：检查最后活动时间）
     const now = new Date()
     const lastActivity = new Date(adminSession.lastActivity || adminSession.loginTime)
@@ -1112,9 +1756,13 @@ const requestLogger = (req, res, next) => {
   const referer = req.get('Referer') || 'none'
 
   // 记录请求开始
+  const isDebugRoute = req.originalUrl.includes('event_logging')
   if (req.originalUrl !== '/health') {
-    // 避免健康检查日志过多
-    logger.info(`▶️ [${requestId}] ${req.method} ${req.originalUrl} | IP: ${clientIP}`)
+    if (isDebugRoute) {
+      logger.debug(`▶️ [${requestId}] ${req.method} ${req.originalUrl} | IP: ${clientIP}`)
+    } else {
+      logger.info(`▶️ [${requestId}] ${req.method} ${req.originalUrl} | IP: ${clientIP}`)
+    }
   }
 
   res.on('finish', () => {
@@ -1146,7 +1794,14 @@ const requestLogger = (req, res, next) => {
         logMetadata
       )
     } else if (req.originalUrl !== '/health') {
-      logger.request(req.method, req.originalUrl, res.statusCode, duration, logMetadata)
+      if (isDebugRoute) {
+        logger.debug(
+          `🟢 ${req.method} ${req.originalUrl} - ${res.statusCode} (${duration}ms)`,
+          logMetadata
+        )
+      } else {
+        logger.request(req.method, req.originalUrl, res.statusCode, duration, logMetadata)
+      }
     }
 
     // API Key相关日志
@@ -1388,7 +2043,8 @@ const globalRateLimit = async (req, res, next) =>
 
 // 📊 请求大小限制中间件
 const requestSizeLimit = (req, res, next) => {
-  const maxSize = 60 * 1024 * 1024 // 60MB
+  const MAX_SIZE_MB = parseInt(process.env.REQUEST_MAX_SIZE_MB || '60', 10)
+  const maxSize = MAX_SIZE_MB * 1024 * 1024
   const contentLength = parseInt(req.headers['content-length'] || '0')
 
   if (contentLength > maxSize) {

src/models/redis.js

@@ -50,6 +50,18 @@ function getWeekStringInTimezone(date = new Date()) {
   return `${year}-W${String(weekNumber).padStart(2, '0')}`
 }
 
+// 并发队列相关常量
+const QUEUE_STATS_TTL_SECONDS = 86400 * 7 // 统计计数保留 7 天
+const WAIT_TIME_TTL_SECONDS = 86400 // 等待时间样本保留 1 天（滚动窗口，无需长期保留）
+// 等待时间样本数配置（提高统计置信度）
+// - 每 API Key 从 100 提高到 500：提供更稳定的 P99 估计
+// - 全局从 500 提高到 2000：支持更高精度的 P99.9 分析
+// - 内存开销约 12-20KB（Redis quicklist 每元素 1-10 字节），可接受
+// 详见 design.md Decision 5: 等待时间统计样本数
+const WAIT_TIME_SAMPLES_PER_KEY = 500 // 每个 API Key 保留的等待时间样本数
+const WAIT_TIME_SAMPLES_GLOBAL = 2000 // 全局保留的等待时间样本数
+const QUEUE_TTL_BUFFER_SECONDS = 30 // 排队计数器TTL缓冲时间
+
 class RedisClient {
   constructor() {
     this.client = null
@@ -84,7 +96,25 @@ class RedisClient {
         logger.warn('⚠️  Redis connection closed')
       })
 
-      await this.client.connect()
+      // 只有在 lazyConnect 模式下才需要手动调用 connect()
+      // 如果 Redis 已经连接或正在连接中，则跳过
+      if (
+        this.client.status !== 'connecting' &&
+        this.client.status !== 'connect' &&
+        this.client.status !== 'ready'
+      ) {
+        await this.client.connect()
+      } else {
+        // 等待 ready 状态
+        await new Promise((resolve, reject) => {
+          if (this.client.status === 'ready') {
+            resolve()
+          } else {
+            this.client.once('ready', resolve)
+            this.client.once('error', reject)
+          }
+        })
+      }
       return this.client
     } catch (error) {
       logger.error('💥 Failed to connect to Redis:', error)
@@ -2110,6 +2140,27 @@ class RedisClient {
       const results = []
 
       for (const key of keys) {
+        // 跳过已知非 Sorted Set 类型的键
+        // - concurrency:queue:stats:* 是 Hash 类型
+        // - concurrency:queue:wait_times:* 是 List 类型
+        // - concurrency:queue:* (不含stats/wait_times) 是 String 类型
+        if (
+          key.startsWith('concurrency:queue:stats:') ||
+          key.startsWith('concurrency:queue:wait_times:') ||
+          (key.startsWith('concurrency:queue:') &&
+            !key.includes(':stats:') &&
+            !key.includes(':wait_times:'))
+        ) {
+          continue
+        }
+
+        // 检查键类型，只处理 Sorted Set
+        const keyType = await client.type(key)
+        if (keyType !== 'zset') {
+          logger.debug(`🔢 getAllConcurrencyStatus skipped non-zset key: ${key} (type: ${keyType})`)
+          continue
+        }
+
         // 提取 apiKeyId（去掉 concurrency: 前缀）
         const apiKeyId = key.replace('concurrency:', '')
 
@@ -2172,6 +2223,23 @@ class RedisClient {
         }
       }
 
+      // 检查键类型，只处理 Sorted Set
+      const keyType = await client.type(key)
+      if (keyType !== 'zset') {
+        logger.warn(
+          `⚠️ getConcurrencyStatus: key ${key} has unexpected type: ${keyType}, expected zset`
+        )
+        return {
+          apiKeyId,
+          key,
+          activeCount: 0,
+          expiredCount: 0,
+          activeRequests: [],
+          exists: true,
+          invalidType: keyType
+        }
+      }
+
       // 获取所有成员和分数
       const allMembers = await client.zrange(key, 0, -1, 'WITHSCORES')
 
@@ -2221,20 +2289,36 @@ class RedisClient {
       const client = this.getClientSafe()
       const key = `concurrency:${apiKeyId}`
 
-      // 获取清理前的状态
-      const beforeCount = await client.zcard(key)
+      // 检查键类型
+      const keyType = await client.type(key)
 
-      // 删除整个 key
+      let beforeCount = 0
+      let isLegacy = false
+
+      if (keyType === 'zset') {
+        // 正常的 zset 键，获取条目数
+        beforeCount = await client.zcard(key)
+      } else if (keyType !== 'none') {
+        // 非 zset 且非空的遗留键
+        isLegacy = true
+        logger.warn(
+          `⚠️ forceClearConcurrency: key ${key} has unexpected type: ${keyType}, will be deleted`
+        )
+      }
+
+      // 删除键（无论什么类型）
       await client.del(key)
 
       logger.warn(
-        `🧹 Force cleared concurrency for key ${apiKeyId}, removed ${beforeCount} entries`
+        `🧹 Force cleared concurrency for key ${apiKeyId}, removed ${beforeCount} entries${isLegacy ? ' (legacy key)' : ''}`
       )
 
       return {
         apiKeyId,
         key,
         clearedCount: beforeCount,
+        type: keyType,
+        legacy: isLegacy,
         success: true
       }
     } catch (error) {
@@ -2253,25 +2337,47 @@ class RedisClient {
       const keys = await client.keys('concurrency:*')
 
       let totalCleared = 0
+      let legacyCleared = 0
       const clearedKeys = []
 
       for (const key of keys) {
-        const count = await client.zcard(key)
-        await client.del(key)
-        totalCleared += count
-        clearedKeys.push({
-          key,
-          clearedCount: count
-        })
+        // 跳过 queue 相关的键（它们有各自的清理逻辑）
+        if (key.startsWith('concurrency:queue:')) {
+          continue
+        }
+
+        // 检查键类型
+        const keyType = await client.type(key)
+        if (keyType === 'zset') {
+          const count = await client.zcard(key)
+          await client.del(key)
+          totalCleared += count
+          clearedKeys.push({
+            key,
+            clearedCount: count,
+            type: 'zset'
+          })
+        } else {
+          // 非 zset 类型的遗留键，直接删除
+          await client.del(key)
+          legacyCleared++
+          clearedKeys.push({
+            key,
+            clearedCount: 0,
+            type: keyType,
+            legacy: true
+          })
+        }
       }
 
       logger.warn(
-        `🧹 Force cleared all concurrency: ${keys.length} keys, ${totalCleared} total entries`
+        `🧹 Force cleared all concurrency: ${clearedKeys.length} keys, ${totalCleared} entries, ${legacyCleared} legacy keys`
       )
 
       return {
-        keysCleared: keys.length,
+        keysCleared: clearedKeys.length,
         totalEntriesCleared: totalCleared,
+        legacyKeysCleared: legacyCleared,
         clearedKeys,
         success: true
       }
@@ -2299,9 +2405,30 @@ class RedisClient {
       }
 
       let totalCleaned = 0
+      let legacyCleaned = 0
       const cleanedKeys = []
 
       for (const key of keys) {
+        // 跳过 queue 相关的键（它们有各自的清理逻辑）
+        if (key.startsWith('concurrency:queue:')) {
+          continue
+        }
+
+        // 检查键类型
+        const keyType = await client.type(key)
+        if (keyType !== 'zset') {
+          // 非 zset 类型的遗留键，直接删除
+          await client.del(key)
+          legacyCleaned++
+          cleanedKeys.push({
+            key,
+            cleanedCount: 0,
+            type: keyType,
+            legacy: true
+          })
+          continue
+        }
+
         // 只清理过期的条目
         const cleaned = await client.zremrangebyscore(key, '-inf', now)
         if (cleaned > 0) {
@@ -2320,13 +2447,14 @@ class RedisClient {
       }
 
       logger.info(
-        `🧹 Cleaned up expired concurrency: ${totalCleaned} entries from ${cleanedKeys.length} keys`
+        `🧹 Cleaned up expired concurrency: ${totalCleaned} entries from ${cleanedKeys.length} keys, ${legacyCleaned} legacy keys removed`
       )
 
       return {
         keysProcessed: keys.length,
         keysCleaned: cleanedKeys.length,
         totalEntriesCleaned: totalCleaned,
+        legacyKeysRemoved: legacyCleaned,
         cleanedKeys,
         success: true
       }
@@ -2769,4 +2897,625 @@ redisClient.scanUserMessageQueueLocks = async function () {
   }
 }
 
+// ============================================
+// 🚦 API Key 并发请求排队方法
+// ============================================
+
+/**
+ * 增加排队计数（使用 Lua 脚本确保原子性）
+ * @param {string} apiKeyId - API Key ID
+ * @param {number} [timeoutMs=60000] - 排队超时时间（毫秒），用于计算 TTL
+ * @returns {Promise<number>} 增加后的排队数量
+ */
+redisClient.incrConcurrencyQueue = async function (apiKeyId, timeoutMs = 60000) {
+  const key = `concurrency:queue:${apiKeyId}`
+  try {
+    // 使用 Lua 脚本确保 INCR 和 EXPIRE 原子执行，防止进程崩溃导致计数器泄漏
+    // TTL = 超时时间 + 缓冲时间（确保键不会在请求还在等待时过期）
+    const ttlSeconds = Math.ceil(timeoutMs / 1000) + QUEUE_TTL_BUFFER_SECONDS
+    const script = `
+      local count = redis.call('INCR', KEYS[1])
+      redis.call('EXPIRE', KEYS[1], ARGV[1])
+      return count
+    `
+    const count = await this.client.eval(script, 1, key, String(ttlSeconds))
+    logger.database(
+      `🚦 Incremented queue count for key ${apiKeyId}: ${count} (TTL: ${ttlSeconds}s)`
+    )
+    return parseInt(count)
+  } catch (error) {
+    logger.error(`Failed to increment concurrency queue for ${apiKeyId}:`, error)
+    throw error
+  }
+}
+
+/**
+ * 减少排队计数（使用 Lua 脚本确保原子性）
+ * @param {string} apiKeyId - API Key ID
+ * @returns {Promise<number>} 减少后的排队数量
+ */
+redisClient.decrConcurrencyQueue = async function (apiKeyId) {
+  const key = `concurrency:queue:${apiKeyId}`
+  try {
+    // 使用 Lua 脚本确保 DECR 和 DEL 原子执行，防止进程崩溃导致计数器残留
+    const script = `
+      local count = redis.call('DECR', KEYS[1])
+      if count <= 0 then
+        redis.call('DEL', KEYS[1])
+        return 0
+      end
+      return count
+    `
+    const count = await this.client.eval(script, 1, key)
+    const result = parseInt(count)
+    if (result === 0) {
+      logger.database(`🚦 Queue count for key ${apiKeyId} is 0, removed key`)
+    } else {
+      logger.database(`🚦 Decremented queue count for key ${apiKeyId}: ${result}`)
+    }
+    return result
+  } catch (error) {
+    logger.error(`Failed to decrement concurrency queue for ${apiKeyId}:`, error)
+    throw error
+  }
+}
+
+/**
+ * 获取排队计数
+ * @param {string} apiKeyId - API Key ID
+ * @returns {Promise<number>} 当前排队数量
+ */
+redisClient.getConcurrencyQueueCount = async function (apiKeyId) {
+  const key = `concurrency:queue:${apiKeyId}`
+  try {
+    const count = await this.client.get(key)
+    return parseInt(count || 0)
+  } catch (error) {
+    logger.error(`Failed to get concurrency queue count for ${apiKeyId}:`, error)
+    return 0
+  }
+}
+
+/**
+ * 清空排队计数
+ * @param {string} apiKeyId - API Key ID
+ * @returns {Promise<boolean>} 是否成功清空
+ */
+redisClient.clearConcurrencyQueue = async function (apiKeyId) {
+  const key = `concurrency:queue:${apiKeyId}`
+  try {
+    await this.client.del(key)
+    logger.database(`🚦 Cleared queue count for key ${apiKeyId}`)
+    return true
+  } catch (error) {
+    logger.error(`Failed to clear concurrency queue for ${apiKeyId}:`, error)
+    return false
+  }
+}
+
+/**
+ * 扫描所有排队计数器
+ * @returns {Promise<string[]>} API Key ID 列表
+ */
+redisClient.scanConcurrencyQueueKeys = async function () {
+  const apiKeyIds = []
+  let cursor = '0'
+  let iterations = 0
+  const MAX_ITERATIONS = 1000
+
+  try {
+    do {
+      const [newCursor, keys] = await this.client.scan(
+        cursor,
+        'MATCH',
+        'concurrency:queue:*',
+        'COUNT',
+        100
+      )
+      cursor = newCursor
+      iterations++
+
+      for (const key of keys) {
+        // 排除统计和等待时间相关的键
+        if (
+          key.startsWith('concurrency:queue:stats:') ||
+          key.startsWith('concurrency:queue:wait_times:')
+        ) {
+          continue
+        }
+        const apiKeyId = key.replace('concurrency:queue:', '')
+        apiKeyIds.push(apiKeyId)
+      }
+
+      if (iterations >= MAX_ITERATIONS) {
+        logger.warn(
+          `🚦 Concurrency queue: SCAN reached max iterations (${MAX_ITERATIONS}), stopping early`,
+          { foundQueues: apiKeyIds.length }
+        )
+        break
+      }
+    } while (cursor !== '0')
+
+    return apiKeyIds
+  } catch (error) {
+    logger.error('Failed to scan concurrency queue keys:', error)
+    return []
+  }
+}
+
+/**
+ * 清理所有排队计数器（用于服务重启）
+ * @returns {Promise<number>} 清理的计数器数量
+ */
+redisClient.clearAllConcurrencyQueues = async function () {
+  let cleared = 0
+  let cursor = '0'
+  let iterations = 0
+  const MAX_ITERATIONS = 1000
+
+  try {
+    do {
+      const [newCursor, keys] = await this.client.scan(
+        cursor,
+        'MATCH',
+        'concurrency:queue:*',
+        'COUNT',
+        100
+      )
+      cursor = newCursor
+      iterations++
+
+      // 只删除排队计数器，保留统计数据
+      const queueKeys = keys.filter(
+        (key) =>
+          !key.startsWith('concurrency:queue:stats:') &&
+          !key.startsWith('concurrency:queue:wait_times:')
+      )
+
+      if (queueKeys.length > 0) {
+        await this.client.del(...queueKeys)
+        cleared += queueKeys.length
+      }
+
+      if (iterations >= MAX_ITERATIONS) {
+        break
+      }
+    } while (cursor !== '0')
+
+    if (cleared > 0) {
+      logger.info(`🚦 Cleared ${cleared} concurrency queue counter(s) on startup`)
+    }
+    return cleared
+  } catch (error) {
+    logger.error('Failed to clear all concurrency queues:', error)
+    return 0
+  }
+}
+
+/**
+ * 增加排队统计计数（使用 Lua 脚本确保原子性）
+ * @param {string} apiKeyId - API Key ID
+ * @param {string} field - 统计字段 (entered/success/timeout/cancelled)
+ * @returns {Promise<number>} 增加后的计数
+ */
+redisClient.incrConcurrencyQueueStats = async function (apiKeyId, field) {
+  const key = `concurrency:queue:stats:${apiKeyId}`
+  try {
+    // 使用 Lua 脚本确保 HINCRBY 和 EXPIRE 原子执行
+    // 防止在两者之间崩溃导致统计键没有 TTL（内存泄漏）
+    const script = `
+      local count = redis.call('HINCRBY', KEYS[1], ARGV[1], 1)
+      redis.call('EXPIRE', KEYS[1], ARGV[2])
+      return count
+    `
+    const count = await this.client.eval(script, 1, key, field, String(QUEUE_STATS_TTL_SECONDS))
+    return parseInt(count)
+  } catch (error) {
+    logger.error(`Failed to increment queue stats ${field} for ${apiKeyId}:`, error)
+    return 0
+  }
+}
+
+/**
+ * 获取排队统计
+ * @param {string} apiKeyId - API Key ID
+ * @returns {Promise<Object>} 统计数据
+ */
+redisClient.getConcurrencyQueueStats = async function (apiKeyId) {
+  const key = `concurrency:queue:stats:${apiKeyId}`
+  try {
+    const stats = await this.client.hgetall(key)
+    return {
+      entered: parseInt(stats?.entered || 0),
+      success: parseInt(stats?.success || 0),
+      timeout: parseInt(stats?.timeout || 0),
+      cancelled: parseInt(stats?.cancelled || 0),
+      socket_changed: parseInt(stats?.socket_changed || 0),
+      rejected_overload: parseInt(stats?.rejected_overload || 0)
+    }
+  } catch (error) {
+    logger.error(`Failed to get queue stats for ${apiKeyId}:`, error)
+    return {
+      entered: 0,
+      success: 0,
+      timeout: 0,
+      cancelled: 0,
+      socket_changed: 0,
+      rejected_overload: 0
+    }
+  }
+}
+
+/**
+ * 记录排队等待时间（按 API Key 分开存储）
+ * @param {string} apiKeyId - API Key ID
+ * @param {number} waitTimeMs - 等待时间（毫秒）
+ * @returns {Promise<void>}
+ */
+redisClient.recordQueueWaitTime = async function (apiKeyId, waitTimeMs) {
+  const key = `concurrency:queue:wait_times:${apiKeyId}`
+  try {
+    // 使用 Lua 脚本确保原子性，同时设置 TTL 防止内存泄漏
+    const script = `
+      redis.call('LPUSH', KEYS[1], ARGV[1])
+      redis.call('LTRIM', KEYS[1], 0, ARGV[2])
+      redis.call('EXPIRE', KEYS[1], ARGV[3])
+      return 1
+    `
+    await this.client.eval(
+      script,
+      1,
+      key,
+      waitTimeMs,
+      WAIT_TIME_SAMPLES_PER_KEY - 1,
+      WAIT_TIME_TTL_SECONDS
+    )
+  } catch (error) {
+    logger.error(`Failed to record queue wait time for ${apiKeyId}:`, error)
+  }
+}
+
+/**
+ * 记录全局排队等待时间
+ * @param {number} waitTimeMs - 等待时间（毫秒）
+ * @returns {Promise<void>}
+ */
+redisClient.recordGlobalQueueWaitTime = async function (waitTimeMs) {
+  const key = 'concurrency:queue:wait_times:global'
+  try {
+    // 使用 Lua 脚本确保原子性，同时设置 TTL 防止内存泄漏
+    const script = `
+      redis.call('LPUSH', KEYS[1], ARGV[1])
+      redis.call('LTRIM', KEYS[1], 0, ARGV[2])
+      redis.call('EXPIRE', KEYS[1], ARGV[3])
+      return 1
+    `
+    await this.client.eval(
+      script,
+      1,
+      key,
+      waitTimeMs,
+      WAIT_TIME_SAMPLES_GLOBAL - 1,
+      WAIT_TIME_TTL_SECONDS
+    )
+  } catch (error) {
+    logger.error('Failed to record global queue wait time:', error)
+  }
+}
+
+/**
+ * 获取全局等待时间列表
+ * @returns {Promise<number[]>} 等待时间列表
+ */
+redisClient.getGlobalQueueWaitTimes = async function () {
+  const key = 'concurrency:queue:wait_times:global'
+  try {
+    const samples = await this.client.lrange(key, 0, -1)
+    return samples.map(Number)
+  } catch (error) {
+    logger.error('Failed to get global queue wait times:', error)
+    return []
+  }
+}
+
+/**
+ * 获取指定 API Key 的等待时间列表
+ * @param {string} apiKeyId - API Key ID
+ * @returns {Promise<number[]>} 等待时间列表
+ */
+redisClient.getQueueWaitTimes = async function (apiKeyId) {
+  const key = `concurrency:queue:wait_times:${apiKeyId}`
+  try {
+    const samples = await this.client.lrange(key, 0, -1)
+    return samples.map(Number)
+  } catch (error) {
+    logger.error(`Failed to get queue wait times for ${apiKeyId}:`, error)
+    return []
+  }
+}
+
+/**
+ * 扫描所有排队统计键
+ * @returns {Promise<string[]>} API Key ID 列表
+ */
+redisClient.scanConcurrencyQueueStatsKeys = async function () {
+  const apiKeyIds = []
+  let cursor = '0'
+  let iterations = 0
+  const MAX_ITERATIONS = 1000
+
+  try {
+    do {
+      const [newCursor, keys] = await this.client.scan(
+        cursor,
+        'MATCH',
+        'concurrency:queue:stats:*',
+        'COUNT',
+        100
+      )
+      cursor = newCursor
+      iterations++
+
+      for (const key of keys) {
+        const apiKeyId = key.replace('concurrency:queue:stats:', '')
+        apiKeyIds.push(apiKeyId)
+      }
+
+      if (iterations >= MAX_ITERATIONS) {
+        break
+      }
+    } while (cursor !== '0')
+
+    return apiKeyIds
+  } catch (error) {
+    logger.error('Failed to scan concurrency queue stats keys:', error)
+    return []
+  }
+}
+
+// ============================================================================
+// 账户测试历史相关操作
+// ============================================================================
+
+const ACCOUNT_TEST_HISTORY_MAX = 5 // 保留最近5次测试记录
+const ACCOUNT_TEST_HISTORY_TTL = 86400 * 30 // 30天过期
+const ACCOUNT_TEST_CONFIG_TTL = 86400 * 365 // 测试配置保留1年（用户通常长期使用）
+
+/**
+ * 保存账户测试结果
+ * @param {string} accountId - 账户ID
+ * @param {string} platform - 平台类型 (claude/gemini/openai等)
+ * @param {Object} testResult - 测试结果对象
+ * @param {boolean} testResult.success - 是否成功
+ * @param {string} testResult.message - 测试消息/响应
+ * @param {number} testResult.latencyMs - 延迟毫秒数
+ * @param {string} testResult.error - 错误信息（如有）
+ * @param {string} testResult.timestamp - 测试时间戳
+ */
+redisClient.saveAccountTestResult = async function (accountId, platform, testResult) {
+  const key = `account:test_history:${platform}:${accountId}`
+  try {
+    const record = JSON.stringify({
+      ...testResult,
+      timestamp: testResult.timestamp || new Date().toISOString()
+    })
+
+    // 使用 LPUSH + LTRIM 保持最近5条记录
+    const client = this.getClientSafe()
+    await client.lpush(key, record)
+    await client.ltrim(key, 0, ACCOUNT_TEST_HISTORY_MAX - 1)
+    await client.expire(key, ACCOUNT_TEST_HISTORY_TTL)
+
+    logger.debug(`📝 Saved test result for ${platform} account ${accountId}`)
+  } catch (error) {
+    logger.error(`Failed to save test result for ${accountId}:`, error)
+  }
+}
+
+/**
+ * 获取账户测试历史
+ * @param {string} accountId - 账户ID
+ * @param {string} platform - 平台类型
+ * @returns {Promise<Array>} 测试历史记录数组（最新在前）
+ */
+redisClient.getAccountTestHistory = async function (accountId, platform) {
+  const key = `account:test_history:${platform}:${accountId}`
+  try {
+    const client = this.getClientSafe()
+    const records = await client.lrange(key, 0, -1)
+    return records.map((r) => JSON.parse(r))
+  } catch (error) {
+    logger.error(`Failed to get test history for ${accountId}:`, error)
+    return []
+  }
+}
+
+/**
+ * 获取账户最新测试结果
+ * @param {string} accountId - 账户ID
+ * @param {string} platform - 平台类型
+ * @returns {Promise<Object|null>} 最新测试结果
+ */
+redisClient.getAccountLatestTestResult = async function (accountId, platform) {
+  const key = `account:test_history:${platform}:${accountId}`
+  try {
+    const client = this.getClientSafe()
+    const record = await client.lindex(key, 0)
+    return record ? JSON.parse(record) : null
+  } catch (error) {
+    logger.error(`Failed to get latest test result for ${accountId}:`, error)
+    return null
+  }
+}
+
+/**
+ * 批量获取多个账户的测试历史
+ * @param {Array<{accountId: string, platform: string}>} accounts - 账户列表
+ * @returns {Promise<Object>} 以 accountId 为 key 的测试历史映射
+ */
+redisClient.getAccountsTestHistory = async function (accounts) {
+  const result = {}
+  try {
+    const client = this.getClientSafe()
+    const pipeline = client.pipeline()
+
+    for (const { accountId, platform } of accounts) {
+      const key = `account:test_history:${platform}:${accountId}`
+      pipeline.lrange(key, 0, -1)
+    }
+
+    const responses = await pipeline.exec()
+
+    accounts.forEach(({ accountId }, index) => {
+      const [err, records] = responses[index]
+      if (!err && records) {
+        result[accountId] = records.map((r) => JSON.parse(r))
+      } else {
+        result[accountId] = []
+      }
+    })
+  } catch (error) {
+    logger.error('Failed to get batch test history:', error)
+  }
+  return result
+}
+
+/**
+ * 保存定时测试配置
+ * @param {string} accountId - 账户ID
+ * @param {string} platform - 平台类型
+ * @param {Object} config - 配置对象
+ * @param {boolean} config.enabled - 是否启用定时测试
+ * @param {string} config.cronExpression - Cron 表达式 (如 "0 8 * * *" 表示每天8点)
+ * @param {string} config.model - 测试使用的模型
+ */
+redisClient.saveAccountTestConfig = async function (accountId, platform, testConfig) {
+  const key = `account:test_config:${platform}:${accountId}`
+  try {
+    const client = this.getClientSafe()
+    await client.hset(key, {
+      enabled: testConfig.enabled ? 'true' : 'false',
+      cronExpression: testConfig.cronExpression || '0 8 * * *', // 默认每天早上8点
+      model: testConfig.model || 'claude-sonnet-4-5-20250929', // 默认模型
+      updatedAt: new Date().toISOString()
+    })
+    // 设置过期时间（1年）
+    await client.expire(key, ACCOUNT_TEST_CONFIG_TTL)
+  } catch (error) {
+    logger.error(`Failed to save test config for ${accountId}:`, error)
+  }
+}
+
+/**
+ * 获取定时测试配置
+ * @param {string} accountId - 账户ID
+ * @param {string} platform - 平台类型
+ * @returns {Promise<Object|null>} 配置对象
+ */
+redisClient.getAccountTestConfig = async function (accountId, platform) {
+  const key = `account:test_config:${platform}:${accountId}`
+  try {
+    const client = this.getClientSafe()
+    const testConfig = await client.hgetall(key)
+    if (!testConfig || Object.keys(testConfig).length === 0) {
+      return null
+    }
+    // 向后兼容：如果存在旧的 testHour 字段，转换为 cron 表达式
+    let { cronExpression } = testConfig
+    if (!cronExpression && testConfig.testHour) {
+      const hour = parseInt(testConfig.testHour, 10)
+      cronExpression = `0 ${hour} * * *`
+    }
+    return {
+      enabled: testConfig.enabled === 'true',
+      cronExpression: cronExpression || '0 8 * * *',
+      model: testConfig.model || 'claude-sonnet-4-5-20250929',
+      updatedAt: testConfig.updatedAt
+    }
+  } catch (error) {
+    logger.error(`Failed to get test config for ${accountId}:`, error)
+    return null
+  }
+}
+
+/**
+ * 获取所有启用定时测试的账户
+ * @param {string} platform - 平台类型
+ * @returns {Promise<Array>} 账户ID列表及 cron 配置
+ */
+redisClient.getEnabledTestAccounts = async function (platform) {
+  const accountIds = []
+  let cursor = '0'
+
+  try {
+    const client = this.getClientSafe()
+    do {
+      const [newCursor, keys] = await client.scan(
+        cursor,
+        'MATCH',
+        `account:test_config:${platform}:*`,
+        'COUNT',
+        100
+      )
+      cursor = newCursor
+
+      for (const key of keys) {
+        const testConfig = await client.hgetall(key)
+        if (testConfig && testConfig.enabled === 'true') {
+          const accountId = key.replace(`account:test_config:${platform}:`, '')
+          // 向后兼容：如果存在旧的 testHour 字段，转换为 cron 表达式
+          let { cronExpression } = testConfig
+          if (!cronExpression && testConfig.testHour) {
+            const hour = parseInt(testConfig.testHour, 10)
+            cronExpression = `0 ${hour} * * *`
+          }
+          accountIds.push({
+            accountId,
+            cronExpression: cronExpression || '0 8 * * *',
+            model: testConfig.model || 'claude-sonnet-4-5-20250929'
+          })
+        }
+      }
+    } while (cursor !== '0')
+
+    return accountIds
+  } catch (error) {
+    logger.error(`Failed to get enabled test accounts for ${platform}:`, error)
+    return []
+  }
+}
+
+/**
+ * 保存账户上次测试时间（用于调度器判断是否需要测试）
+ * @param {string} accountId - 账户ID
+ * @param {string} platform - 平台类型
+ */
+redisClient.setAccountLastTestTime = async function (accountId, platform) {
+  const key = `account:last_test:${platform}:${accountId}`
+  try {
+    const client = this.getClientSafe()
+    await client.set(key, Date.now().toString(), 'EX', 86400 * 7) // 7天过期
+  } catch (error) {
+    logger.error(`Failed to set last test time for ${accountId}:`, error)
+  }
+}
+
+/**
+ * 获取账户上次测试时间
+ * @param {string} accountId - 账户ID
+ * @param {string} platform - 平台类型
+ * @returns {Promise<number|null>} 上次测试时间戳
+ */
+redisClient.getAccountLastTestTime = async function (accountId, platform) {
+  const key = `account:last_test:${platform}:${accountId}`
+  try {
+    const client = this.getClientSafe()
+    const timestamp = await client.get(key)
+    return timestamp ? parseInt(timestamp, 10) : null
+  } catch (error) {
+    logger.error(`Failed to get last test time for ${accountId}:`, error)
+    return null
+  }
+}
+
 module.exports = redisClient

src/routes/admin/apiKeys.js

@@ -945,6 +945,30 @@ async function calculateKeyStats(keyId, timeRange, startDate, endDate) {
       allTimeCost = parseFloat((await client.get(totalCostKey)) || '0')
     }
 
+    // 🔧 FIX: 对于 "全部时间" 时间范围，直接使用 allTimeCost
+    // 因为 usage:*:model:daily:* 键有 30 天 TTL，旧数据已经过期
+    if (timeRange === 'all' && allTimeCost > 0) {
+      logger.debug(`📊 使用 allTimeCost 计算 timeRange='all': ${allTimeCost}`)
+
+      return {
+        requests: 0, // 旧数据详情不可用
+        tokens: 0,
+        inputTokens: 0,
+        outputTokens: 0,
+        cacheCreateTokens: 0,
+        cacheReadTokens: 0,
+        cost: allTimeCost,
+        formattedCost: CostCalculator.formatCost(allTimeCost),
+        // 实时限制数据（始终返回，不受时间范围影响）
+        dailyCost,
+        currentWindowCost,
+        windowRemainingSeconds,
+        windowStartTime,
+        windowEndTime,
+        allTimeCost
+      }
+    }
+
     // 只在启用了窗口限制时查询窗口数据
     if (rateLimitWindow > 0) {
       const costCountKey = `rate_limit:cost:${keyId}`
@@ -1006,12 +1030,10 @@ async function calculateKeyStats(keyId, timeRange, startDate, endDate) {
   const modelStatsMap = new Map()
   let totalRequests = 0
 
-  // 用于去重：只统计日数据，避免与月数据重复
+  // 用于去重：先统计月数据，避免与日数据重复
   const dailyKeyPattern = /usage:.+:model:daily:(.+):\d{4}-\d{2}-\d{2}$/
   const monthlyKeyPattern = /usage:.+:model:monthly:(.+):\d{4}-\d{2}$/
-
-  // 检查是否有日数据
-  const hasDailyData = uniqueKeys.some((key) => dailyKeyPattern.test(key))
+  const currentMonth = `${tzDate.getUTCFullYear()}-${String(tzDate.getUTCMonth() + 1).padStart(2, '0')}`
 
   for (let i = 0; i < results.length; i++) {
     const [err, data] = results[i]
@@ -1038,8 +1060,12 @@ async function calculateKeyStats(keyId, timeRange, startDate, endDate) {
       continue
     }
 
-    // 如果有日数据，则跳过月数据以避免重复
-    if (hasDailyData && isMonthly) {
+    // 跳过当前月的月数据
+    if (isMonthly && key.includes(`:${currentMonth}`)) {
+      continue
+    }
+    // 跳过非当前月的日数据
+    if (!isMonthly && !key.includes(`:${currentMonth}-`)) {
       continue
     }
 

src/routes/admin/claudeAccounts.js

@@ -9,6 +9,7 @@ const router = express.Router()
 const claudeAccountService = require('../../services/claudeAccountService')
 const claudeRelayService = require('../../services/claudeRelayService')
 const accountGroupService = require('../../services/accountGroupService')
+const accountTestSchedulerService = require('../../services/accountTestSchedulerService')
 const apiKeyService = require('../../services/apiKeyService')
 const redis = require('../../models/redis')
 const { authenticateAdmin } = require('../../middleware/auth')
@@ -277,7 +278,7 @@ router.post('/claude-accounts/oauth-with-cookie', authenticateAdmin, async (req,
 
     logger.info('🍪 Starting Cookie-based OAuth authorization', {
       sessionKeyLength: trimmedSessionKey.length,
-      sessionKeyPrefix: trimmedSessionKey.substring(0, 10) + '...',
+      sessionKeyPrefix: `${trimmedSessionKey.substring(0, 10)}...`,
       hasProxy: !!proxy
     })
 
@@ -326,7 +327,7 @@ router.post('/claude-accounts/setup-token-with-cookie', authenticateAdmin, async
 
     logger.info('🍪 Starting Cookie-based Setup Token authorization', {
       sessionKeyLength: trimmedSessionKey.length,
-      sessionKeyPrefix: trimmedSessionKey.substring(0, 10) + '...',
+      sessionKeyPrefix: `${trimmedSessionKey.substring(0, 10)}...`,
       hasProxy: !!proxy
     })
 
@@ -583,7 +584,9 @@ router.post('/claude-accounts', authenticateAdmin, async (req, res) => {
       useUnifiedClientId,
       unifiedClientId,
       expiresAt,
-      extInfo
+      extInfo,
+      maxConcurrency,
+      interceptWarmup
     } = req.body
 
     if (!name) {
@@ -628,7 +631,9 @@ router.post('/claude-accounts', authenticateAdmin, async (req, res) => {
       useUnifiedClientId: useUnifiedClientId === true, // 默认为false
       unifiedClientId: unifiedClientId || '', // 统一的客户端标识
       expiresAt: expiresAt || null, // 账户订阅到期时间
-      extInfo: extInfo || null
+      extInfo: extInfo || null,
+      maxConcurrency: maxConcurrency || 0, // 账户级串行队列：0=使用全局配置，>0=强制启用
+      interceptWarmup: interceptWarmup === true // 拦截预热请求：默认为false
     })
 
     // 如果是分组类型，将账户添加到分组
@@ -903,4 +908,219 @@ router.post('/claude-accounts/:accountId/test', authenticateAdmin, async (req, r
   }
 })
 
+// ============================================================================
+// 账户定时测试相关端点
+// ============================================================================
+
+// 获取账户测试历史
+router.get('/claude-accounts/:accountId/test-history', authenticateAdmin, async (req, res) => {
+  const { accountId } = req.params
+
+  try {
+    const history = await redis.getAccountTestHistory(accountId, 'claude')
+    return res.json({
+      success: true,
+      data: {
+        accountId,
+        platform: 'claude',
+        history
+      }
+    })
+  } catch (error) {
+    logger.error(`❌ Failed to get test history for account ${accountId}:`, error)
+    return res.status(500).json({
+      error: 'Failed to get test history',
+      message: error.message
+    })
+  }
+})
+
+// 获取账户定时测试配置
+router.get('/claude-accounts/:accountId/test-config', authenticateAdmin, async (req, res) => {
+  const { accountId } = req.params
+
+  try {
+    const testConfig = await redis.getAccountTestConfig(accountId, 'claude')
+    return res.json({
+      success: true,
+      data: {
+        accountId,
+        platform: 'claude',
+        config: testConfig || {
+          enabled: false,
+          cronExpression: '0 8 * * *',
+          model: 'claude-sonnet-4-5-20250929'
+        }
+      }
+    })
+  } catch (error) {
+    logger.error(`❌ Failed to get test config for account ${accountId}:`, error)
+    return res.status(500).json({
+      error: 'Failed to get test config',
+      message: error.message
+    })
+  }
+})
+
+// 设置账户定时测试配置
+router.put('/claude-accounts/:accountId/test-config', authenticateAdmin, async (req, res) => {
+  const { accountId } = req.params
+  const { enabled, cronExpression, model } = req.body
+
+  try {
+    // 验证 enabled 参数
+    if (typeof enabled !== 'boolean') {
+      return res.status(400).json({
+        error: 'Invalid parameter',
+        message: 'enabled must be a boolean'
+      })
+    }
+
+    // 验证 cronExpression 参数
+    if (!cronExpression || typeof cronExpression !== 'string') {
+      return res.status(400).json({
+        error: 'Invalid parameter',
+        message: 'cronExpression is required and must be a string'
+      })
+    }
+
+    // 限制 cronExpression 长度防止 DoS
+    const MAX_CRON_LENGTH = 100
+    if (cronExpression.length > MAX_CRON_LENGTH) {
+      return res.status(400).json({
+        error: 'Invalid parameter',
+        message: `cronExpression too long (max ${MAX_CRON_LENGTH} characters)`
+      })
+    }
+
+    // 使用 service 的方法验证 cron 表达式
+    if (!accountTestSchedulerService.validateCronExpression(cronExpression)) {
+      return res.status(400).json({
+        error: 'Invalid parameter',
+        message: `Invalid cron expression: ${cronExpression}. Format: "minute hour day month weekday" (e.g., "0 8 * * *" for daily at 8:00)`
+      })
+    }
+
+    // 验证模型参数
+    const testModel = model || 'claude-sonnet-4-5-20250929'
+    if (typeof testModel !== 'string' || testModel.length > 256) {
+      return res.status(400).json({
+        error: 'Invalid parameter',
+        message: 'model must be a valid string (max 256 characters)'
+      })
+    }
+
+    // 检查账户是否存在
+    const account = await claudeAccountService.getAccount(accountId)
+    if (!account) {
+      return res.status(404).json({
+        error: 'Account not found',
+        message: `Claude account ${accountId} not found`
+      })
+    }
+
+    // 保存配置
+    await redis.saveAccountTestConfig(accountId, 'claude', {
+      enabled,
+      cronExpression,
+      model: testModel
+    })
+
+    logger.success(
+      `📝 Updated test config for Claude account ${accountId}: enabled=${enabled}, cronExpression=${cronExpression}, model=${testModel}`
+    )
+
+    return res.json({
+      success: true,
+      message: 'Test config updated successfully',
+      data: {
+        accountId,
+        platform: 'claude',
+        config: { enabled, cronExpression, model: testModel }
+      }
+    })
+  } catch (error) {
+    logger.error(`❌ Failed to update test config for account ${accountId}:`, error)
+    return res.status(500).json({
+      error: 'Failed to update test config',
+      message: error.message
+    })
+  }
+})
+
+// 手动触发账户测试（非流式，返回JSON结果）
+router.post('/claude-accounts/:accountId/test-sync', authenticateAdmin, async (req, res) => {
+  const { accountId } = req.params
+
+  try {
+    // 检查账户是否存在
+    const account = await claudeAccountService.getAccount(accountId)
+    if (!account) {
+      return res.status(404).json({
+        error: 'Account not found',
+        message: `Claude account ${accountId} not found`
+      })
+    }
+
+    logger.info(`🧪 Manual sync test triggered for Claude account: ${accountId}`)
+
+    // 执行测试
+    const testResult = await claudeRelayService.testAccountConnectionSync(accountId)
+
+    // 保存测试结果到历史
+    await redis.saveAccountTestResult(accountId, 'claude', testResult)
+    await redis.setAccountLastTestTime(accountId, 'claude')
+
+    return res.json({
+      success: true,
+      data: {
+        accountId,
+        platform: 'claude',
+        result: testResult
+      }
+    })
+  } catch (error) {
+    logger.error(`❌ Failed to run sync test for account ${accountId}:`, error)
+    return res.status(500).json({
+      error: 'Failed to run test',
+      message: error.message
+    })
+  }
+})
+
+// 批量获取多个账户的测试历史
+router.post('/claude-accounts/batch-test-history', authenticateAdmin, async (req, res) => {
+  const { accountIds } = req.body
+
+  try {
+    if (!Array.isArray(accountIds) || accountIds.length === 0) {
+      return res.status(400).json({
+        error: 'Invalid parameter',
+        message: 'accountIds must be a non-empty array'
+      })
+    }
+
+    // 限制批量查询数量
+    const limitedIds = accountIds.slice(0, 100)
+
+    const accounts = limitedIds.map((accountId) => ({
+      accountId,
+      platform: 'claude'
+    }))
+
+    const historyMap = await redis.getAccountsTestHistory(accounts)
+
+    return res.json({
+      success: true,
+      data: historyMap
+    })
+  } catch (error) {
+    logger.error('❌ Failed to get batch test history:', error)
+    return res.status(500).json({
+      error: 'Failed to get batch test history',
+      message: error.message
+    })
+  }
+})
+
 module.exports = router

src/routes/admin/claudeConsoleAccounts.js

@@ -132,7 +132,8 @@ router.post('/claude-console-accounts', authenticateAdmin, async (req, res) => {
       dailyQuota,
       quotaResetTime,
       maxConcurrentTasks,
-      disableAutoProtection
+      disableAutoProtection,
+      interceptWarmup
     } = req.body
 
     if (!name || !apiUrl || !apiKey) {
@@ -186,7 +187,8 @@ router.post('/claude-console-accounts', authenticateAdmin, async (req, res) => {
         maxConcurrentTasks !== undefined && maxConcurrentTasks !== null
           ? Number(maxConcurrentTasks)
           : 0,
-      disableAutoProtection: normalizedDisableAutoProtection
+      disableAutoProtection: normalizedDisableAutoProtection,
+      interceptWarmup: interceptWarmup === true || interceptWarmup === 'true'
     })
 
     // 如果是分组类型，将账户添加到分组（CCR 归属 Claude 平台分组）

src/routes/admin/claudeRelayConfig.js

@@ -43,7 +43,11 @@ router.put('/claude-relay-config', authenticateAdmin, async (req, res) => {
       sessionBindingTtlDays,
       userMessageQueueEnabled,
       userMessageQueueDelayMs,
-      userMessageQueueTimeoutMs
+      userMessageQueueTimeoutMs,
+      concurrentRequestQueueEnabled,
+      concurrentRequestQueueMaxSize,
+      concurrentRequestQueueMaxSizeMultiplier,
+      concurrentRequestQueueTimeoutMs
     } = req.body
 
     // 验证输入
@@ -110,6 +114,54 @@ router.put('/claude-relay-config', authenticateAdmin, async (req, res) => {
       }
     }
 
+    // 验证并发请求排队配置
+    if (
+      concurrentRequestQueueEnabled !== undefined &&
+      typeof concurrentRequestQueueEnabled !== 'boolean'
+    ) {
+      return res.status(400).json({ error: 'concurrentRequestQueueEnabled must be a boolean' })
+    }
+
+    if (concurrentRequestQueueMaxSize !== undefined) {
+      if (
+        typeof concurrentRequestQueueMaxSize !== 'number' ||
+        !Number.isInteger(concurrentRequestQueueMaxSize) ||
+        concurrentRequestQueueMaxSize < 1 ||
+        concurrentRequestQueueMaxSize > 100
+      ) {
+        return res
+          .status(400)
+          .json({ error: 'concurrentRequestQueueMaxSize must be an integer between 1 and 100' })
+      }
+    }
+
+    if (concurrentRequestQueueMaxSizeMultiplier !== undefined) {
+      // 使用 Number.isFinite() 同时排除 NaN、Infinity、-Infinity 和非数字类型
+      if (
+        !Number.isFinite(concurrentRequestQueueMaxSizeMultiplier) ||
+        concurrentRequestQueueMaxSizeMultiplier < 0 ||
+        concurrentRequestQueueMaxSizeMultiplier > 10
+      ) {
+        return res.status(400).json({
+          error: 'concurrentRequestQueueMaxSizeMultiplier must be a finite number between 0 and 10'
+        })
+      }
+    }
+
+    if (concurrentRequestQueueTimeoutMs !== undefined) {
+      if (
+        typeof concurrentRequestQueueTimeoutMs !== 'number' ||
+        !Number.isInteger(concurrentRequestQueueTimeoutMs) ||
+        concurrentRequestQueueTimeoutMs < 5000 ||
+        concurrentRequestQueueTimeoutMs > 300000
+      ) {
+        return res.status(400).json({
+          error:
+            'concurrentRequestQueueTimeoutMs must be an integer between 5000 and 300000 (5 seconds to 5 minutes)'
+        })
+      }
+    }
+
     const updateData = {}
     if (claudeCodeOnlyEnabled !== undefined) {
       updateData.claudeCodeOnlyEnabled = claudeCodeOnlyEnabled
@@ -132,6 +184,18 @@ router.put('/claude-relay-config', authenticateAdmin, async (req, res) => {
     if (userMessageQueueTimeoutMs !== undefined) {
       updateData.userMessageQueueTimeoutMs = userMessageQueueTimeoutMs
     }
+    if (concurrentRequestQueueEnabled !== undefined) {
+      updateData.concurrentRequestQueueEnabled = concurrentRequestQueueEnabled
+    }
+    if (concurrentRequestQueueMaxSize !== undefined) {
+      updateData.concurrentRequestQueueMaxSize = concurrentRequestQueueMaxSize
+    }
+    if (concurrentRequestQueueMaxSizeMultiplier !== undefined) {
+      updateData.concurrentRequestQueueMaxSizeMultiplier = concurrentRequestQueueMaxSizeMultiplier
+    }
+    if (concurrentRequestQueueTimeoutMs !== undefined) {
+      updateData.concurrentRequestQueueTimeoutMs = concurrentRequestQueueTimeoutMs
+    }
 
     const updatedConfig = await claudeRelayConfigService.updateConfig(
       updateData,

src/routes/admin/concurrency.js

@@ -8,6 +8,7 @@ const router = express.Router()
 const redis = require('../../models/redis')
 const logger = require('../../utils/logger')
 const { authenticateAdmin } = require('../../middleware/auth')
+const { calculateWaitTimeStats } = require('../../utils/statsHelper')
 
 /**
  * GET /admin/concurrency
@@ -17,17 +18,29 @@ router.get('/concurrency', authenticateAdmin, async (req, res) => {
   try {
     const status = await redis.getAllConcurrencyStatus()
 
+    // 为每个 API Key 获取排队计数
+    const statusWithQueue = await Promise.all(
+      status.map(async (s) => {
+        const queueCount = await redis.getConcurrencyQueueCount(s.apiKeyId)
+        return {
+          ...s,
+          queueCount
+        }
+      })
+    )
+
     // 计算汇总统计
     const summary = {
-      totalKeys: status.length,
-      totalActiveRequests: status.reduce((sum, s) => sum + s.activeCount, 0),
-      totalExpiredRequests: status.reduce((sum, s) => sum + s.expiredCount, 0)
+      totalKeys: statusWithQueue.length,
+      totalActiveRequests: statusWithQueue.reduce((sum, s) => sum + s.activeCount, 0),
+      totalExpiredRequests: statusWithQueue.reduce((sum, s) => sum + s.expiredCount, 0),
+      totalQueuedRequests: statusWithQueue.reduce((sum, s) => sum + s.queueCount, 0)
     }
 
     res.json({
       success: true,
       summary,
-      concurrencyStatus: status
+      concurrencyStatus: statusWithQueue
     })
   } catch (error) {
     logger.error('❌ Failed to get concurrency status:', error)
@@ -39,6 +52,156 @@ router.get('/concurrency', authenticateAdmin, async (req, res) => {
   }
 })
 
+/**
+ * GET /admin/concurrency-queue/stats
+ * 获取排队统计信息
+ */
+router.get('/concurrency-queue/stats', authenticateAdmin, async (req, res) => {
+  try {
+    // 获取所有有统计数据的 API Key
+    const statsKeys = await redis.scanConcurrencyQueueStatsKeys()
+    const queueKeys = await redis.scanConcurrencyQueueKeys()
+
+    // 合并所有相关的 API Key
+    const allApiKeyIds = [...new Set([...statsKeys, ...queueKeys])]
+
+    // 获取各 API Key 的详细统计
+    const perKeyStats = await Promise.all(
+      allApiKeyIds.map(async (apiKeyId) => {
+        const [queueCount, stats, waitTimes] = await Promise.all([
+          redis.getConcurrencyQueueCount(apiKeyId),
+          redis.getConcurrencyQueueStats(apiKeyId),
+          redis.getQueueWaitTimes(apiKeyId)
+        ])
+
+        return {
+          apiKeyId,
+          currentQueueCount: queueCount,
+          stats,
+          waitTimeStats: calculateWaitTimeStats(waitTimes)
+        }
+      })
+    )
+
+    // 获取全局等待时间统计
+    const globalWaitTimes = await redis.getGlobalQueueWaitTimes()
+    const globalWaitTimeStats = calculateWaitTimeStats(globalWaitTimes)
+
+    // 计算全局汇总
+    const globalStats = {
+      totalEntered: perKeyStats.reduce((sum, s) => sum + s.stats.entered, 0),
+      totalSuccess: perKeyStats.reduce((sum, s) => sum + s.stats.success, 0),
+      totalTimeout: perKeyStats.reduce((sum, s) => sum + s.stats.timeout, 0),
+      totalCancelled: perKeyStats.reduce((sum, s) => sum + s.stats.cancelled, 0),
+      totalSocketChanged: perKeyStats.reduce((sum, s) => sum + (s.stats.socket_changed || 0), 0),
+      totalRejectedOverload: perKeyStats.reduce(
+        (sum, s) => sum + (s.stats.rejected_overload || 0),
+        0
+      ),
+      currentTotalQueued: perKeyStats.reduce((sum, s) => sum + s.currentQueueCount, 0),
+      // 队列资源利用率指标
+      peakQueueSize:
+        perKeyStats.length > 0 ? Math.max(...perKeyStats.map((s) => s.currentQueueCount)) : 0,
+      avgQueueSize:
+        perKeyStats.length > 0
+          ? Math.round(
+              perKeyStats.reduce((sum, s) => sum + s.currentQueueCount, 0) / perKeyStats.length
+            )
+          : 0,
+      activeApiKeys: perKeyStats.filter((s) => s.currentQueueCount > 0).length
+    }
+
+    // 计算成功率
+    if (globalStats.totalEntered > 0) {
+      globalStats.successRate = Math.round(
+        (globalStats.totalSuccess / globalStats.totalEntered) * 100
+      )
+      globalStats.timeoutRate = Math.round(
+        (globalStats.totalTimeout / globalStats.totalEntered) * 100
+      )
+      globalStats.cancelledRate = Math.round(
+        (globalStats.totalCancelled / globalStats.totalEntered) * 100
+      )
+    }
+
+    // 从全局等待时间统计中提取关键指标
+    if (globalWaitTimeStats) {
+      globalStats.avgWaitTimeMs = globalWaitTimeStats.avg
+      globalStats.p50WaitTimeMs = globalWaitTimeStats.p50
+      globalStats.p90WaitTimeMs = globalWaitTimeStats.p90
+      globalStats.p99WaitTimeMs = globalWaitTimeStats.p99
+      // 多实例采样策略标记（详见 design.md Decision 9）
+      // 全局 P90 仅用于可视化和监控，不用于系统决策
+      // 健康检查使用 API Key 级别的 P90（每 Key 独立采样）
+      globalWaitTimeStats.globalP90ForVisualizationOnly = true
+    }
+
+    res.json({
+      success: true,
+      globalStats,
+      globalWaitTimeStats,
+      perKeyStats
+    })
+  } catch (error) {
+    logger.error('❌ Failed to get queue stats:', error)
+    res.status(500).json({
+      success: false,
+      error: 'Failed to get queue stats',
+      message: error.message
+    })
+  }
+})
+
+/**
+ * DELETE /admin/concurrency-queue/:apiKeyId
+ * 清理特定 API Key 的排队计数
+ */
+router.delete('/concurrency-queue/:apiKeyId', authenticateAdmin, async (req, res) => {
+  try {
+    const { apiKeyId } = req.params
+    await redis.clearConcurrencyQueue(apiKeyId)
+
+    logger.warn(`🧹 Admin ${req.admin?.username || 'unknown'} cleared queue for key ${apiKeyId}`)
+
+    res.json({
+      success: true,
+      message: `Successfully cleared queue for API key ${apiKeyId}`
+    })
+  } catch (error) {
+    logger.error(`❌ Failed to clear queue for ${req.params.apiKeyId}:`, error)
+    res.status(500).json({
+      success: false,
+      error: 'Failed to clear queue',
+      message: error.message
+    })
+  }
+})
+
+/**
+ * DELETE /admin/concurrency-queue
+ * 清理所有排队计数
+ */
+router.delete('/concurrency-queue', authenticateAdmin, async (req, res) => {
+  try {
+    const cleared = await redis.clearAllConcurrencyQueues()
+
+    logger.warn(`🧹 Admin ${req.admin?.username || 'unknown'} cleared ALL queues`)
+
+    res.json({
+      success: true,
+      message: 'Successfully cleared all queues',
+      cleared
+    })
+  } catch (error) {
+    logger.error('❌ Failed to clear all queues:', error)
+    res.status(500).json({
+      success: false,
+      error: 'Failed to clear all queues',
+      message: error.message
+    })
+  }
+})
+
 /**
  * GET /admin/concurrency/:apiKeyId
  * 获取特定 API Key 的并发状态详情
@@ -47,10 +210,14 @@ router.get('/concurrency/:apiKeyId', authenticateAdmin, async (req, res) => {
   try {
     const { apiKeyId } = req.params
     const status = await redis.getConcurrencyStatus(apiKeyId)
+    const queueCount = await redis.getConcurrencyQueueCount(apiKeyId)
 
     res.json({
       success: true,
-      concurrencyStatus: status
+      concurrencyStatus: {
+        ...status,
+        queueCount
+      }
     })
   } catch (error) {
     logger.error(`❌ Failed to get concurrency status for ${req.params.apiKeyId}:`, error)

src/routes/admin/index.js

@@ -24,6 +24,7 @@ const usageStatsRoutes = require('./usageStats')
 const systemRoutes = require('./system')
 const concurrencyRoutes = require('./concurrency')
 const claudeRelayConfigRoutes = require('./claudeRelayConfig')
+const syncRoutes = require('./sync')
 
 // 挂载所有子路由
 // 使用完整路径的模块（直接挂载到根路径）
@@ -39,6 +40,7 @@ router.use('/', usageStatsRoutes)
 router.use('/', systemRoutes)
 router.use('/', concurrencyRoutes)
 router.use('/', claudeRelayConfigRoutes)
+router.use('/', syncRoutes)
 
 // 使用相对路径的模块（需要指定基础路径前缀）
 router.use('/account-groups', accountGroupsRoutes)

src/routes/admin/sync.js

@@ -0,0 +1,460 @@
+/**
+ * Admin Routes - Sync / Export (for migration)
+ * Exports account data (including secrets) for safe server-to-server syncing.
+ */
+
+const express = require('express')
+const router = express.Router()
+
+const { authenticateAdmin } = require('../../middleware/auth')
+const redis = require('../../models/redis')
+const claudeAccountService = require('../../services/claudeAccountService')
+const claudeConsoleAccountService = require('../../services/claudeConsoleAccountService')
+const openaiAccountService = require('../../services/openaiAccountService')
+const openaiResponsesAccountService = require('../../services/openaiResponsesAccountService')
+const logger = require('../../utils/logger')
+
+function toBool(value, defaultValue = false) {
+  if (value === undefined || value === null || value === '') {
+    return defaultValue
+  }
+  if (value === true || value === 'true') {
+    return true
+  }
+  if (value === false || value === 'false') {
+    return false
+  }
+  return defaultValue
+}
+
+function normalizeProxy(proxy) {
+  if (!proxy || typeof proxy !== 'object') {
+    return null
+  }
+
+  const protocol = proxy.protocol || proxy.type || proxy.scheme || ''
+  const host = proxy.host || ''
+  const port = Number(proxy.port || 0)
+
+  if (!protocol || !host || !Number.isFinite(port) || port <= 0) {
+    return null
+  }
+
+  return {
+    protocol: String(protocol),
+    host: String(host),
+    port,
+    username: proxy.username ? String(proxy.username) : '',
+    password: proxy.password ? String(proxy.password) : ''
+  }
+}
+
+function buildModelMappingFromSupportedModels(supportedModels) {
+  if (!supportedModels) {
+    return null
+  }
+
+  if (Array.isArray(supportedModels)) {
+    const mapping = {}
+    for (const model of supportedModels) {
+      if (typeof model === 'string' && model.trim()) {
+        mapping[model.trim()] = model.trim()
+      }
+    }
+    return Object.keys(mapping).length ? mapping : null
+  }
+
+  if (typeof supportedModels === 'object') {
+    const mapping = {}
+    for (const [from, to] of Object.entries(supportedModels)) {
+      if (typeof from === 'string' && typeof to === 'string' && from.trim() && to.trim()) {
+        mapping[from.trim()] = to.trim()
+      }
+    }
+    return Object.keys(mapping).length ? mapping : null
+  }
+
+  return null
+}
+
+function safeParseJson(raw, fallback = null) {
+  if (!raw || typeof raw !== 'string') {
+    return fallback
+  }
+  try {
+    return JSON.parse(raw)
+  } catch (_) {
+    return fallback
+  }
+}
+
+// Export accounts for migration (includes secrets).
+// GET /admin/sync/export-accounts?include_secrets=true
+router.get('/sync/export-accounts', authenticateAdmin, async (req, res) => {
+  try {
+    const includeSecrets = toBool(req.query.include_secrets, false)
+    if (!includeSecrets) {
+      return res.status(400).json({
+        success: false,
+        error: 'include_secrets_required',
+        message: 'Set include_secrets=true to export secrets'
+      })
+    }
+
+    // ===== Claude official OAuth / Setup Token accounts =====
+    const rawClaudeAccounts = await redis.getAllClaudeAccounts()
+    const claudeAccounts = rawClaudeAccounts.map((account) => {
+      // Backward compatible extraction: prefer individual fields, fallback to claudeAiOauth JSON blob.
+      let decryptedClaudeAiOauth = null
+      if (account.claudeAiOauth) {
+        try {
+          const raw = claudeAccountService._decryptSensitiveData(account.claudeAiOauth)
+          decryptedClaudeAiOauth = raw ? JSON.parse(raw) : null
+        } catch (_) {
+          decryptedClaudeAiOauth = null
+        }
+      }
+
+      const rawScopes =
+        account.scopes && account.scopes.trim()
+          ? account.scopes
+          : decryptedClaudeAiOauth?.scopes
+            ? decryptedClaudeAiOauth.scopes.join(' ')
+            : ''
+
+      const scopes = rawScopes && rawScopes.trim() ? rawScopes.trim().split(' ') : []
+      const isOAuth = scopes.includes('user:profile') && scopes.includes('user:inference')
+      const authType = isOAuth ? 'oauth' : 'setup-token'
+
+      const accessToken =
+        account.accessToken && String(account.accessToken).trim()
+          ? claudeAccountService._decryptSensitiveData(account.accessToken)
+          : decryptedClaudeAiOauth?.accessToken || ''
+
+      const refreshToken =
+        account.refreshToken && String(account.refreshToken).trim()
+          ? claudeAccountService._decryptSensitiveData(account.refreshToken)
+          : decryptedClaudeAiOauth?.refreshToken || ''
+
+      let expiresAt = null
+      const expiresAtMs = Number.parseInt(account.expiresAt, 10)
+      if (Number.isFinite(expiresAtMs) && expiresAtMs > 0) {
+        expiresAt = new Date(expiresAtMs).toISOString()
+      } else if (decryptedClaudeAiOauth?.expiresAt) {
+        try {
+          expiresAt = new Date(Number(decryptedClaudeAiOauth.expiresAt)).toISOString()
+        } catch (_) {
+          expiresAt = null
+        }
+      }
+
+      const proxy = account.proxy ? normalizeProxy(safeParseJson(account.proxy)) : null
+
+      // 🔧 Parse subscriptionInfo to extract org_uuid and account_uuid
+      let orgUuid = null
+      let accountUuid = null
+      if (account.subscriptionInfo) {
+        try {
+          const subscriptionInfo = JSON.parse(account.subscriptionInfo)
+          orgUuid = subscriptionInfo.organizationUuid || null
+          accountUuid = subscriptionInfo.accountUuid || null
+        } catch (_) {
+          // Ignore parse errors
+        }
+      }
+
+      // 🔧 Calculate expires_in from expires_at
+      let expiresIn = null
+      if (expiresAt) {
+        try {
+          const expiresAtTime = new Date(expiresAt).getTime()
+          const nowTime = Date.now()
+          const diffSeconds = Math.floor((expiresAtTime - nowTime) / 1000)
+          if (diffSeconds > 0) {
+            expiresIn = diffSeconds
+          }
+        } catch (_) {
+          // Ignore calculation errors
+        }
+      }
+      // 🔧 Use default expires_in if calculation failed (Anthropic OAuth: 8 hours)
+      if (!expiresIn && isOAuth) {
+        expiresIn = 28800 // 8 hours
+      }
+
+      const credentials = {
+        access_token: accessToken,
+        refresh_token: refreshToken || undefined,
+        expires_at: expiresAt || undefined,
+        expires_in: expiresIn || undefined,
+        scope: scopes.join(' ') || undefined,
+        token_type: 'Bearer'
+      }
+      // 🔧 Add auth info as top-level credentials fields
+      if (orgUuid) {
+        credentials.org_uuid = orgUuid
+      }
+      if (accountUuid) {
+        credentials.account_uuid = accountUuid
+      }
+
+      // 🔧 Store complete original CRS data in extra
+      const extra = {
+        crs_account_id: account.id,
+        crs_kind: 'claude-account',
+        crs_id: account.id,
+        crs_name: account.name,
+        crs_description: account.description || '',
+        crs_platform: account.platform || 'claude',
+        crs_auth_type: authType,
+        crs_is_active: account.isActive === 'true',
+        crs_schedulable: account.schedulable !== 'false',
+        crs_priority: Number.parseInt(account.priority, 10) || 50,
+        crs_status: account.status || 'active',
+        crs_scopes: scopes,
+        crs_subscription_info: account.subscriptionInfo || undefined
+      }
+
+      return {
+        kind: 'claude-account',
+        id: account.id,
+        name: account.name,
+        description: account.description || '',
+        platform: account.platform || 'claude',
+        authType,
+        isActive: account.isActive === 'true',
+        schedulable: account.schedulable !== 'false',
+        priority: Number.parseInt(account.priority, 10) || 50,
+        status: account.status || 'active',
+        proxy,
+        credentials,
+        extra
+      }
+    })
+
+    // ===== Claude Console API Key accounts =====
+    const claudeConsoleSummaries = await claudeConsoleAccountService.getAllAccounts()
+    const claudeConsoleAccounts = []
+    for (const summary of claudeConsoleSummaries) {
+      const full = await claudeConsoleAccountService.getAccount(summary.id)
+      if (!full) {
+        continue
+      }
+
+      const proxy = normalizeProxy(full.proxy)
+      const modelMapping = buildModelMappingFromSupportedModels(full.supportedModels)
+
+      const credentials = {
+        api_key: full.apiKey,
+        base_url: full.apiUrl
+      }
+
+      if (modelMapping) {
+        credentials.model_mapping = modelMapping
+      }
+
+      if (full.userAgent) {
+        credentials.user_agent = full.userAgent
+      }
+
+      claudeConsoleAccounts.push({
+        kind: 'claude-console-account',
+        id: full.id,
+        name: full.name,
+        description: full.description || '',
+        platform: full.platform || 'claude-console',
+        isActive: full.isActive === true,
+        schedulable: full.schedulable !== false,
+        priority: Number.parseInt(full.priority, 10) || 50,
+        status: full.status || 'active',
+        proxy,
+        maxConcurrentTasks: Number.parseInt(full.maxConcurrentTasks, 10) || 0,
+        credentials,
+        extra: {
+          crs_account_id: full.id,
+          crs_kind: 'claude-console-account',
+          crs_id: full.id,
+          crs_name: full.name,
+          crs_description: full.description || '',
+          crs_platform: full.platform || 'claude-console',
+          crs_is_active: full.isActive === true,
+          crs_schedulable: full.schedulable !== false,
+          crs_priority: Number.parseInt(full.priority, 10) || 50,
+          crs_status: full.status || 'active'
+        }
+      })
+    }
+
+    // ===== OpenAI OAuth accounts =====
+    const openaiOAuthAccounts = []
+    {
+      const client = redis.getClientSafe()
+      const openaiKeys = await client.keys('openai:account:*')
+      for (const key of openaiKeys) {
+        const id = key.split(':').slice(2).join(':')
+        const account = await openaiAccountService.getAccount(id)
+        if (!account) {
+          continue
+        }
+
+        const accessToken = account.accessToken
+          ? openaiAccountService.decrypt(account.accessToken)
+          : ''
+        if (!accessToken) {
+          // Skip broken/legacy records without decryptable token
+          continue
+        }
+
+        const scopes =
+          account.scopes && typeof account.scopes === 'string' && account.scopes.trim()
+            ? account.scopes.trim().split(' ')
+            : []
+
+        const proxy = normalizeProxy(account.proxy)
+
+        // 🔧 Calculate expires_in from expires_at
+        let expiresIn = null
+        if (account.expiresAt) {
+          try {
+            const expiresAtTime = new Date(account.expiresAt).getTime()
+            const nowTime = Date.now()
+            const diffSeconds = Math.floor((expiresAtTime - nowTime) / 1000)
+            if (diffSeconds > 0) {
+              expiresIn = diffSeconds
+            }
+          } catch (_) {
+            // Ignore calculation errors
+          }
+        }
+        // 🔧 Use default expires_in if calculation failed (OpenAI OAuth: 10 days)
+        if (!expiresIn) {
+          expiresIn = 864000 // 10 days
+        }
+
+        const credentials = {
+          access_token: accessToken,
+          refresh_token: account.refreshToken || undefined,
+          id_token: account.idToken || undefined,
+          expires_at: account.expiresAt || undefined,
+          expires_in: expiresIn || undefined,
+          scope: scopes.join(' ') || undefined,
+          token_type: 'Bearer'
+        }
+        // 🔧 Add auth info as top-level credentials fields
+        if (account.accountId) {
+          credentials.chatgpt_account_id = account.accountId
+        }
+        if (account.chatgptUserId) {
+          credentials.chatgpt_user_id = account.chatgptUserId
+        }
+        if (account.organizationId) {
+          credentials.organization_id = account.organizationId
+        }
+
+        // 🔧 Store complete original CRS data in extra
+        const extra = {
+          crs_account_id: account.id,
+          crs_kind: 'openai-oauth-account',
+          crs_id: account.id,
+          crs_name: account.name,
+          crs_description: account.description || '',
+          crs_platform: account.platform || 'openai',
+          crs_is_active: account.isActive === 'true',
+          crs_schedulable: account.schedulable !== 'false',
+          crs_priority: Number.parseInt(account.priority, 10) || 50,
+          crs_status: account.status || 'active',
+          crs_scopes: scopes,
+          crs_email: account.email || undefined,
+          crs_chatgpt_account_id: account.accountId || undefined,
+          crs_chatgpt_user_id: account.chatgptUserId || undefined,
+          crs_organization_id: account.organizationId || undefined
+        }
+
+        openaiOAuthAccounts.push({
+          kind: 'openai-oauth-account',
+          id: account.id,
+          name: account.name,
+          description: account.description || '',
+          platform: account.platform || 'openai',
+          authType: 'oauth',
+          isActive: account.isActive === 'true',
+          schedulable: account.schedulable !== 'false',
+          priority: Number.parseInt(account.priority, 10) || 50,
+          status: account.status || 'active',
+          proxy,
+          credentials,
+          extra
+        })
+      }
+    }
+
+    // ===== OpenAI Responses API Key accounts =====
+    const openaiResponsesAccounts = []
+    const client = redis.getClientSafe()
+    const openaiResponseKeys = await client.keys('openai_responses_account:*')
+    for (const key of openaiResponseKeys) {
+      const id = key.split(':').slice(1).join(':')
+      const full = await openaiResponsesAccountService.getAccount(id)
+      if (!full) {
+        continue
+      }
+
+      const proxy = normalizeProxy(full.proxy)
+
+      const credentials = {
+        api_key: full.apiKey,
+        base_url: full.baseApi
+      }
+
+      if (full.userAgent) {
+        credentials.user_agent = full.userAgent
+      }
+
+      openaiResponsesAccounts.push({
+        kind: 'openai-responses-account',
+        id: full.id,
+        name: full.name,
+        description: full.description || '',
+        platform: full.platform || 'openai-responses',
+        isActive: full.isActive === 'true',
+        schedulable: full.schedulable !== 'false',
+        priority: Number.parseInt(full.priority, 10) || 50,
+        status: full.status || 'active',
+        proxy,
+        credentials,
+        extra: {
+          crs_account_id: full.id,
+          crs_kind: 'openai-responses-account',
+          crs_id: full.id,
+          crs_name: full.name,
+          crs_description: full.description || '',
+          crs_platform: full.platform || 'openai-responses',
+          crs_is_active: full.isActive === 'true',
+          crs_schedulable: full.schedulable !== 'false',
+          crs_priority: Number.parseInt(full.priority, 10) || 50,
+          crs_status: full.status || 'active'
+        }
+      })
+    }
+
+    return res.json({
+      success: true,
+      data: {
+        exportedAt: new Date().toISOString(),
+        claudeAccounts,
+        claudeConsoleAccounts,
+        openaiOAuthAccounts,
+        openaiResponsesAccounts
+      }
+    })
+  } catch (error) {
+    logger.error('❌ Failed to export accounts for sync:', error)
+    return res.status(500).json({
+      success: false,
+      error: 'export_failed',
+      message: error.message
+    })
+  }
+})
+
+module.exports = router

src/routes/api.js

@@ -12,6 +12,13 @@ const { getEffectiveModel, parseVendorPrefixedModel } = require('../utils/modelH
 const sessionHelper = require('../utils/sessionHelper')
 const { updateRateLimitCounters } = require('../utils/rateLimitHelper')
 const claudeRelayConfigService = require('../services/claudeRelayConfigService')
+const claudeAccountService = require('../services/claudeAccountService')
+const claudeConsoleAccountService = require('../services/claudeConsoleAccountService')
+const {
+  isWarmupRequest,
+  buildMockWarmupResponse,
+  sendMockWarmupStream
+} = require('../utils/warmupInterceptor')
 const { sanitizeUpstreamError } = require('../utils/errorSanitizer')
 const router = express.Router()
 
@@ -190,12 +197,42 @@ async function handleMessagesRequest(req, res) {
     )
 
     if (isStream) {
+      // 🔍 检查客户端连接是否仍然有效（可能在并发排队等待期间断开）
+      if (res.destroyed || res.socket?.destroyed || res.writableEnded) {
+        logger.warn(
+          `⚠️ Client disconnected before stream response could start for key: ${req.apiKey?.name || 'unknown'}`
+        )
+        return undefined
+      }
+
       // 流式响应 - 只使用官方真实usage数据
       res.setHeader('Content-Type', 'text/event-stream')
       res.setHeader('Cache-Control', 'no-cache')
       res.setHeader('Connection', 'keep-alive')
       res.setHeader('Access-Control-Allow-Origin', '*')
       res.setHeader('X-Accel-Buffering', 'no') // 禁用 Nginx 缓冲
+      // ⚠️ 检查 headers 是否已发送（可能在排队心跳时已设置）
+      if (!res.headersSent) {
+        res.setHeader('Content-Type', 'text/event-stream')
+        res.setHeader('Cache-Control', 'no-cache')
+        // ⚠️ 关键修复：尊重 auth.js 提前设置的 Connection: close
+        // 当并发队列功能启用时，auth.js 会设置 Connection: close 来禁用 Keep-Alive
+        // 这里只在没有设置过 Connection 头时才设置 keep-alive
+        const existingConnection = res.getHeader('Connection')
+        if (!existingConnection) {
+          res.setHeader('Connection', 'keep-alive')
+        } else {
+          logger.api(
+            `🔌 [STREAM] Preserving existing Connection header: ${existingConnection} for key: ${req.apiKey?.name || 'unknown'}`
+          )
+        }
+        res.setHeader('Access-Control-Allow-Origin', '*')
+        res.setHeader('X-Accel-Buffering', 'no') // 禁用 Nginx 缓冲
+      } else {
+        logger.debug(
+          `📤 [STREAM] Headers already sent, skipping setHeader for key: ${req.apiKey?.name || 'unknown'}`
+        )
+      }
 
       // 禁用 Nagle 算法，确保数据立即发送
       if (res.socket && typeof res.socket.setNoDelay === 'function') {
@@ -333,6 +370,23 @@ async function handleMessagesRequest(req, res) {
         }
       }
 
+      // 🔥 预热请求拦截检查（在转发之前）
+      if (accountType === 'claude-official' || accountType === 'claude-console') {
+        const account =
+          accountType === 'claude-official'
+            ? await claudeAccountService.getAccount(accountId)
+            : await claudeConsoleAccountService.getAccount(accountId)
+
+        if (account?.interceptWarmup === 'true' && isWarmupRequest(req.body)) {
+          logger.api(`🔥 Warmup request intercepted for account: ${account.name} (${accountId})`)
+          if (isStream) {
+            return sendMockWarmupStream(res, req.body.model)
+          } else {
+            return res.json(buildMockWarmupResponse(req.body.model))
+          }
+        }
+      }
+
       // 根据账号类型选择对应的转发服务并调用
       if (accountType === 'claude-official') {
         // 官方Claude账号使用原有的转发服务（会自己选择账号）
@@ -657,12 +711,61 @@ async function handleMessagesRequest(req, res) {
         }
       }, 1000) // 1秒后检查
     } else {
+      // 🔍 检查客户端连接是否仍然有效（可能在并发排队等待期间断开）
+      if (res.destroyed || res.socket?.destroyed || res.writableEnded) {
+        logger.warn(
+          `⚠️ Client disconnected before non-stream request could start for key: ${req.apiKey?.name || 'unknown'}`
+        )
+        return undefined
+      }
+
       // 非流式响应 - 只使用官方真实usage数据
       logger.info('📄 Starting non-streaming request', {
         apiKeyId: req.apiKey.id,
         apiKeyName: req.apiKey.name
       })
 
+      // 📊 监听 socket 事件以追踪连接状态变化
+      const nonStreamSocket = res.socket
+      let _clientClosedConnection = false
+      let _socketCloseTime = null
+
+      if (nonStreamSocket) {
+        const onSocketEnd = () => {
+          _clientClosedConnection = true
+          _socketCloseTime = Date.now()
+          logger.warn(
+            `⚠️ [NON-STREAM] Socket 'end' event - client sent FIN | key: ${req.apiKey?.name}, ` +
+              `requestId: ${req.requestId}, elapsed: ${Date.now() - startTime}ms`
+          )
+        }
+        const onSocketClose = () => {
+          _clientClosedConnection = true
+          logger.warn(
+            `⚠️ [NON-STREAM] Socket 'close' event | key: ${req.apiKey?.name}, ` +
+              `requestId: ${req.requestId}, elapsed: ${Date.now() - startTime}ms, ` +
+              `hadError: ${nonStreamSocket.destroyed}`
+          )
+        }
+        const onSocketError = (err) => {
+          logger.error(
+            `❌ [NON-STREAM] Socket error | key: ${req.apiKey?.name}, ` +
+              `requestId: ${req.requestId}, error: ${err.message}`
+          )
+        }
+
+        nonStreamSocket.once('end', onSocketEnd)
+        nonStreamSocket.once('close', onSocketClose)
+        nonStreamSocket.once('error', onSocketError)
+
+        // 清理监听器（在响应结束后）
+        res.once('finish', () => {
+          nonStreamSocket.removeListener('end', onSocketEnd)
+          nonStreamSocket.removeListener('close', onSocketClose)
+          nonStreamSocket.removeListener('error', onSocketError)
+        })
+      }
+
       // 生成会话哈希用于sticky会话
       const sessionHash = sessionHelper.generateSessionHash(req.body)
 
@@ -783,6 +886,21 @@ async function handleMessagesRequest(req, res) {
         }
       }
 
+      // 🔥 预热请求拦截检查（非流式，在转发之前）
+      if (accountType === 'claude-official' || accountType === 'claude-console') {
+        const account =
+          accountType === 'claude-official'
+            ? await claudeAccountService.getAccount(accountId)
+            : await claudeConsoleAccountService.getAccount(accountId)
+
+        if (account?.interceptWarmup === 'true' && isWarmupRequest(req.body)) {
+          logger.api(
+            `🔥 Warmup request intercepted (non-stream) for account: ${account.name} (${accountId})`
+          )
+          return res.json(buildMockWarmupResponse(req.body.model))
+        }
+      }
+
       // 根据账号类型选择对应的转发服务
       let response
       logger.debug(`[DEBUG] Request query params: ${JSON.stringify(req.query)}`)
@@ -867,6 +985,15 @@ async function handleMessagesRequest(req, res) {
         bodyLength: response.body ? response.body.length : 0
       })
 
+      // 🔍 检查客户端连接是否仍然有效
+      // 在长时间请求过程中，客户端可能已经断开连接（超时、用户取消等）
+      if (res.destroyed || res.socket?.destroyed || res.writableEnded) {
+        logger.warn(
+          `⚠️ Client disconnected before non-stream response could be sent for key: ${req.apiKey?.name || 'unknown'}`
+        )
+        return undefined
+      }
+
       res.status(response.statusCode)
 
       // 设置响应头，避免 Content-Length 和 Transfer-Encoding 冲突
@@ -932,10 +1059,12 @@ async function handleMessagesRequest(req, res) {
           logger.warn('⚠️ No usage data found in Claude API JSON response')
         }
 
+        // 使用 Express 内建的 res.json() 发送响应（简单可靠）
         res.json(jsonData)
       } catch (parseError) {
         logger.warn('⚠️ Failed to parse Claude API response as JSON:', parseError.message)
         logger.info('📄 Raw response body:', response.body)
+        // 使用 Express 内建的 res.send() 发送响应（简单可靠）
         res.send(response.body)
       }
 
@@ -1264,9 +1393,6 @@ router.post('/v1/messages/count_tokens', authenticateApiKey, async (req, res) =>
   const maxAttempts = 2
   let attempt = 0
 
-  // 引入 claudeConsoleAccountService 用于检查 count_tokens 可用性
-  const claudeConsoleAccountService = require('../services/claudeConsoleAccountService')
-
   const processRequest = async () => {
     const { accountId, accountType } = await unifiedClaudeScheduler.selectAccountForApiKey(
       req.apiKey,
@@ -1462,5 +1588,10 @@ router.post('/v1/messages/count_tokens', authenticateApiKey, async (req, res) =>
   }
 })
 
+// Claude Code 客户端遥测端点 - 返回成功响应避免 404 日志
+router.post('/api/event_logging/batch', (req, res) => {
+  res.status(200).json({ success: true })
+})
+
 module.exports = router
 module.exports.handleMessagesRequest = handleMessagesRequest

src/routes/apiStats.js

@@ -206,74 +206,85 @@ router.post('/api/user-stats', async (req, res) => {
     // 获取验证结果中的完整keyData（包含isActive状态和cost信息）
     const fullKeyData = keyData
 
-    // 计算总费用 - 使用与模型统计相同的逻辑（按模型分别计算）
+    // 🔧 FIX: 使用 allTimeCost 而不是扫描月度键
+    // 计算总费用 - 优先使用持久化的总费用计数器
     let totalCost = 0
     let formattedCost = '$0.000000'
 
     try {
       const client = redis.getClientSafe()
 
-      // 获取所有月度模型统计（与model-stats接口相同的逻辑）
-      const allModelKeys = await client.keys(`usage:${keyId}:model:monthly:*:*`)
-      const modelUsageMap = new Map()
+      // 读取累积的总费用（没有 TTL 的持久键）
+      const totalCostKey = `usage:cost:total:${keyId}`
+      const allTimeCost = parseFloat((await client.get(totalCostKey)) || '0')
 
-      for (const key of allModelKeys) {
-        const modelMatch = key.match(/usage:.+:model:monthly:(.+):(\d{4}-\d{2})$/)
-        if (!modelMatch) {
-          continue
-        }
+      if (allTimeCost > 0) {
+        totalCost = allTimeCost
+        formattedCost = CostCalculator.formatCost(allTimeCost)
+        logger.debug(`📊 使用 allTimeCost 计算用户统计: ${allTimeCost}`)
+      } else {
+        // Fallback: 如果 allTimeCost 为空（旧键），尝试月度键
+        const allModelKeys = await client.keys(`usage:${keyId}:model:monthly:*:*`)
+        const modelUsageMap = new Map()
 
-        const model = modelMatch[1]
-        const data = await client.hgetall(key)
-
-        if (data && Object.keys(data).length > 0) {
-          if (!modelUsageMap.has(model)) {
-            modelUsageMap.set(model, {
-              inputTokens: 0,
-              outputTokens: 0,
-              cacheCreateTokens: 0,
-              cacheReadTokens: 0
-            })
+        for (const key of allModelKeys) {
+          const modelMatch = key.match(/usage:.+:model:monthly:(.+):(\d{4}-\d{2})$/)
+          if (!modelMatch) {
+            continue
           }
 
-          const modelUsage = modelUsageMap.get(model)
-          modelUsage.inputTokens += parseInt(data.inputTokens) || 0
-          modelUsage.outputTokens += parseInt(data.outputTokens) || 0
-          modelUsage.cacheCreateTokens += parseInt(data.cacheCreateTokens) || 0
-          modelUsage.cacheReadTokens += parseInt(data.cacheReadTokens) || 0
-        }
-      }
+          const model = modelMatch[1]
+          const data = await client.hgetall(key)
 
-      // 按模型计算费用并汇总
-      for (const [model, usage] of modelUsageMap) {
-        const usageData = {
-          input_tokens: usage.inputTokens,
-          output_tokens: usage.outputTokens,
-          cache_creation_input_tokens: usage.cacheCreateTokens,
-          cache_read_input_tokens: usage.cacheReadTokens
+          if (data && Object.keys(data).length > 0) {
+            if (!modelUsageMap.has(model)) {
+              modelUsageMap.set(model, {
+                inputTokens: 0,
+                outputTokens: 0,
+                cacheCreateTokens: 0,
+                cacheReadTokens: 0
+              })
+            }
+
+            const modelUsage = modelUsageMap.get(model)
+            modelUsage.inputTokens += parseInt(data.inputTokens) || 0
+            modelUsage.outputTokens += parseInt(data.outputTokens) || 0
+            modelUsage.cacheCreateTokens += parseInt(data.cacheCreateTokens) || 0
+            modelUsage.cacheReadTokens += parseInt(data.cacheReadTokens) || 0
+          }
         }
 
-        const costResult = CostCalculator.calculateCost(usageData, model)
-        totalCost += costResult.costs.total
-      }
+        // 按模型计算费用并汇总
+        for (const [model, usage] of modelUsageMap) {
+          const usageData = {
+            input_tokens: usage.inputTokens,
+            output_tokens: usage.outputTokens,
+            cache_creation_input_tokens: usage.cacheCreateTokens,
+            cache_read_input_tokens: usage.cacheReadTokens
+          }
 
-      // 如果没有模型级别的详细数据，回退到总体数据计算
-      if (modelUsageMap.size === 0 && fullKeyData.usage?.total?.allTokens > 0) {
-        const usage = fullKeyData.usage.total
-        const costUsage = {
-          input_tokens: usage.inputTokens || 0,
-          output_tokens: usage.outputTokens || 0,
-          cache_creation_input_tokens: usage.cacheCreateTokens || 0,
-          cache_read_input_tokens: usage.cacheReadTokens || 0
+          const costResult = CostCalculator.calculateCost(usageData, model)
+          totalCost += costResult.costs.total
         }
 
-        const costResult = CostCalculator.calculateCost(costUsage, 'claude-3-5-sonnet-20241022')
-        totalCost = costResult.costs.total
-      }
+        // 如果没有模型级别的详细数据，回退到总体数据计算
+        if (modelUsageMap.size === 0 && fullKeyData.usage?.total?.allTokens > 0) {
+          const usage = fullKeyData.usage.total
+          const costUsage = {
+            input_tokens: usage.inputTokens || 0,
+            output_tokens: usage.outputTokens || 0,
+            cache_creation_input_tokens: usage.cacheCreateTokens || 0,
+            cache_read_input_tokens: usage.cacheReadTokens || 0
+          }
 
-      formattedCost = CostCalculator.formatCost(totalCost)
+          const costResult = CostCalculator.calculateCost(costUsage, 'claude-3-5-sonnet-20241022')
+          totalCost = costResult.costs.total
+        }
+
+        formattedCost = CostCalculator.formatCost(totalCost)
+      }
     } catch (error) {
-      logger.warn(`Failed to calculate detailed cost for key ${keyId}:`, error)
+      logger.warn(`Failed to calculate cost for key ${keyId}:`, error)
       // 回退到简单计算
       if (fullKeyData.usage?.total?.allTokens > 0) {
         const usage = fullKeyData.usage.total

src/routes/web.js

@@ -164,13 +164,27 @@ router.post('/auth/change-password', async (req, res) => {
 
     // 获取当前会话
     const sessionData = await redis.getSession(token)
-    if (!sessionData) {
+
+    // 🔒 安全修复：检查空对象
+    if (!sessionData || Object.keys(sessionData).length === 0) {
       return res.status(401).json({
         error: 'Invalid token',
         message: 'Session expired or invalid'
       })
     }
 
+    // 🔒 安全修复：验证会话完整性
+    if (!sessionData.username || !sessionData.loginTime) {
+      logger.security(
+        `🔒 Invalid session structure in /auth/change-password from ${req.ip || 'unknown'}`
+      )
+      await redis.deleteSession(token)
+      return res.status(401).json({
+        error: 'Invalid session',
+        message: 'Session data corrupted or incomplete'
+      })
+    }
+
     // 获取当前管理员信息
     const adminData = await redis.getSession('admin_credentials')
     if (!adminData) {
@@ -269,13 +283,25 @@ router.get('/auth/user', async (req, res) => {
 
     // 获取当前会话
     const sessionData = await redis.getSession(token)
-    if (!sessionData) {
+
+    // 🔒 安全修复：检查空对象
+    if (!sessionData || Object.keys(sessionData).length === 0) {
       return res.status(401).json({
         error: 'Invalid token',
         message: 'Session expired or invalid'
       })
     }
 
+    // 🔒 安全修复：验证会话完整性
+    if (!sessionData.username || !sessionData.loginTime) {
+      logger.security(`🔒 Invalid session structure in /auth/user from ${req.ip || 'unknown'}`)
+      await redis.deleteSession(token)
+      return res.status(401).json({
+        error: 'Invalid session',
+        message: 'Session data corrupted or incomplete'
+      })
+    }
+
     // 获取管理员信息
     const adminData = await redis.getSession('admin_credentials')
     if (!adminData) {
@@ -316,13 +342,24 @@ router.post('/auth/refresh', async (req, res) => {
 
     const sessionData = await redis.getSession(token)
 
-    if (!sessionData) {
+    // 🔒 安全修复：检查空对象（hgetall 对不存在的 key 返回 {}）
+    if (!sessionData || Object.keys(sessionData).length === 0) {
       return res.status(401).json({
         error: 'Invalid token',
         message: 'Session expired or invalid'
       })
     }
 
+    // 🔒 安全修复：验证会话完整性（必须有 username 和 loginTime）
+    if (!sessionData.username || !sessionData.loginTime) {
+      logger.security(`🔒 Invalid session structure detected from ${req.ip || 'unknown'}`)
+      await redis.deleteSession(token) // 清理无效/伪造的会话
+      return res.status(401).json({
+        error: 'Invalid session',
+        message: 'Session data corrupted or incomplete'
+      })
+    }
+
     // 更新最后活动时间
     sessionData.lastActivity = new Date().toISOString()
     await redis.setSession(token, sessionData, config.security.adminSessionTimeout)

src/services/accountTestSchedulerService.js

@@ -0,0 +1,420 @@
+/**
+ * 账户定时测试调度服务
+ * 使用 node-cron 支持 crontab 表达式，为每个账户创建独立的定时任务
+ */
+
+const cron = require('node-cron')
+const redis = require('../models/redis')
+const logger = require('../utils/logger')
+
+class AccountTestSchedulerService {
+  constructor() {
+    // 存储每个账户的 cron 任务: Map<string, { task: ScheduledTask, cronExpression: string }>
+    this.scheduledTasks = new Map()
+    // 定期刷新配置的间隔 (毫秒)
+    this.refreshIntervalMs = 60 * 1000
+    this.refreshInterval = null
+    // 当前正在测试的账户
+    this.testingAccounts = new Set()
+    // 是否已启动
+    this.isStarted = false
+  }
+
+  /**
+   * 验证 cron 表达式是否有效
+   * @param {string} cronExpression - cron 表达式
+   * @returns {boolean}
+   */
+  validateCronExpression(cronExpression) {
+    // 长度检查（防止 DoS）
+    if (!cronExpression || cronExpression.length > 100) {
+      return false
+    }
+    return cron.validate(cronExpression)
+  }
+
+  /**
+   * 启动调度器
+   */
+  async start() {
+    if (this.isStarted) {
+      logger.warn('⚠️ Account test scheduler is already running')
+      return
+    }
+
+    this.isStarted = true
+    logger.info('🚀 Starting account test scheduler service (node-cron mode)')
+
+    // 初始化所有已配置账户的定时任务
+    await this._refreshAllTasks()
+
+    // 定期刷新配置，以便动态添加/修改的配置能生效
+    this.refreshInterval = setInterval(() => {
+      this._refreshAllTasks()
+    }, this.refreshIntervalMs)
+
+    logger.info(
+      `📅 Account test scheduler started (refreshing configs every ${this.refreshIntervalMs / 1000}s)`
+    )
+  }
+
+  /**
+   * 停止调度器
+   */
+  stop() {
+    if (this.refreshInterval) {
+      clearInterval(this.refreshInterval)
+      this.refreshInterval = null
+    }
+
+    // 停止所有 cron 任务
+    for (const [accountKey, taskInfo] of this.scheduledTasks.entries()) {
+      taskInfo.task.stop()
+      logger.debug(`🛑 Stopped cron task for ${accountKey}`)
+    }
+    this.scheduledTasks.clear()
+
+    this.isStarted = false
+    logger.info('🛑 Account test scheduler stopped')
+  }
+
+  /**
+   * 刷新所有账户的定时任务
+   * @private
+   */
+  async _refreshAllTasks() {
+    try {
+      const platforms = ['claude', 'gemini', 'openai']
+      const activeAccountKeys = new Set()
+
+      // 并行加载所有平台的配置
+      const allEnabledAccounts = await Promise.all(
+        platforms.map((platform) =>
+          redis
+            .getEnabledTestAccounts(platform)
+            .then((accounts) => accounts.map((acc) => ({ ...acc, platform })))
+            .catch((error) => {
+              logger.warn(`⚠️ Failed to load test accounts for platform ${platform}:`, error)
+              return []
+            })
+        )
+      )
+
+      // 展平平台数据
+      const flatAccounts = allEnabledAccounts.flat()
+
+      for (const { accountId, cronExpression, model, platform } of flatAccounts) {
+        if (!cronExpression) {
+          logger.warn(
+            `⚠️ Account ${accountId} (${platform}) has no valid cron expression, skipping`
+          )
+          continue
+        }
+
+        const accountKey = `${platform}:${accountId}`
+        activeAccountKeys.add(accountKey)
+
+        // 检查是否需要更新任务
+        const existingTask = this.scheduledTasks.get(accountKey)
+        if (existingTask) {
+          // 如果 cron 表达式和模型都没变，不需要更新
+          if (existingTask.cronExpression === cronExpression && existingTask.model === model) {
+            continue
+          }
+          // 配置变了，停止旧任务
+          existingTask.task.stop()
+          logger.info(`🔄 Updating cron task for ${accountKey}: ${cronExpression}, model: ${model}`)
+        } else {
+          logger.info(`➕ Creating cron task for ${accountKey}: ${cronExpression}, model: ${model}`)
+        }
+
+        // 创建新的 cron 任务
+        this._createCronTask(accountId, platform, cronExpression, model)
+      }
+
+      // 清理已删除或禁用的账户任务
+      for (const [accountKey, taskInfo] of this.scheduledTasks.entries()) {
+        if (!activeAccountKeys.has(accountKey)) {
+          taskInfo.task.stop()
+          this.scheduledTasks.delete(accountKey)
+          logger.info(`➖ Removed cron task for ${accountKey} (disabled or deleted)`)
+        }
+      }
+    } catch (error) {
+      logger.error('❌ Error refreshing account test tasks:', error)
+    }
+  }
+
+  /**
+   * 为单个账户创建 cron 任务
+   * @param {string} accountId
+   * @param {string} platform
+   * @param {string} cronExpression
+   * @param {string} model - 测试使用的模型
+   * @private
+   */
+  _createCronTask(accountId, platform, cronExpression, model) {
+    const accountKey = `${platform}:${accountId}`
+
+    // 验证 cron 表达式
+    if (!this.validateCronExpression(cronExpression)) {
+      logger.error(`❌ Invalid cron expression for ${accountKey}: ${cronExpression}`)
+      return
+    }
+
+    const task = cron.schedule(
+      cronExpression,
+      async () => {
+        await this._runAccountTest(accountId, platform, model)
+      },
+      {
+        scheduled: true,
+        timezone: process.env.TZ || 'Asia/Shanghai'
+      }
+    )
+
+    this.scheduledTasks.set(accountKey, {
+      task,
+      cronExpression,
+      model,
+      accountId,
+      platform
+    })
+  }
+
+  /**
+   * 执行单个账户测试
+   * @param {string} accountId - 账户ID
+   * @param {string} platform - 平台类型
+   * @param {string} model - 测试使用的模型
+   * @private
+   */
+  async _runAccountTest(accountId, platform, model) {
+    const accountKey = `${platform}:${accountId}`
+
+    // 避免重复测试
+    if (this.testingAccounts.has(accountKey)) {
+      logger.debug(`⏳ Account ${accountKey} is already being tested, skipping`)
+      return
+    }
+
+    this.testingAccounts.add(accountKey)
+
+    try {
+      logger.info(
+        `🧪 Running scheduled test for ${platform} account: ${accountId} (model: ${model})`
+      )
+
+      let testResult
+
+      // 根据平台调用对应的测试方法
+      switch (platform) {
+        case 'claude':
+          testResult = await this._testClaudeAccount(accountId, model)
+          break
+        case 'gemini':
+          testResult = await this._testGeminiAccount(accountId, model)
+          break
+        case 'openai':
+          testResult = await this._testOpenAIAccount(accountId, model)
+          break
+        default:
+          testResult = {
+            success: false,
+            error: `Unsupported platform: ${platform}`,
+            timestamp: new Date().toISOString()
+          }
+      }
+
+      // 保存测试结果
+      await redis.saveAccountTestResult(accountId, platform, testResult)
+
+      // 更新最后测试时间
+      await redis.setAccountLastTestTime(accountId, platform)
+
+      // 记录日志
+      if (testResult.success) {
+        logger.info(
+          `✅ Scheduled test passed for ${platform} account ${accountId} (${testResult.latencyMs}ms)`
+        )
+      } else {
+        logger.warn(
+          `❌ Scheduled test failed for ${platform} account ${accountId}: ${testResult.error}`
+        )
+      }
+
+      return testResult
+    } catch (error) {
+      logger.error(`❌ Error testing ${platform} account ${accountId}:`, error)
+
+      const errorResult = {
+        success: false,
+        error: error.message,
+        timestamp: new Date().toISOString()
+      }
+
+      await redis.saveAccountTestResult(accountId, platform, errorResult)
+      await redis.setAccountLastTestTime(accountId, platform)
+
+      return errorResult
+    } finally {
+      this.testingAccounts.delete(accountKey)
+    }
+  }
+
+  /**
+   * 测试 Claude 账户
+   * @param {string} accountId
+   * @param {string} model - 测试使用的模型
+   * @private
+   */
+  async _testClaudeAccount(accountId, model) {
+    const claudeRelayService = require('./claudeRelayService')
+    return await claudeRelayService.testAccountConnectionSync(accountId, model)
+  }
+
+  /**
+   * 测试 Gemini 账户
+   * @param {string} _accountId
+   * @param {string} _model
+   * @private
+   */
+  async _testGeminiAccount(_accountId, _model) {
+    // Gemini 测试暂时返回未实现
+    return {
+      success: false,
+      error: 'Gemini scheduled test not implemented yet',
+      timestamp: new Date().toISOString()
+    }
+  }
+
+  /**
+   * 测试 OpenAI 账户
+   * @param {string} _accountId
+   * @param {string} _model
+   * @private
+   */
+  async _testOpenAIAccount(_accountId, _model) {
+    // OpenAI 测试暂时返回未实现
+    return {
+      success: false,
+      error: 'OpenAI scheduled test not implemented yet',
+      timestamp: new Date().toISOString()
+    }
+  }
+
+  /**
+   * 手动触发账户测试
+   * @param {string} accountId - 账户ID
+   * @param {string} platform - 平台类型
+   * @param {string} model - 测试使用的模型
+   * @returns {Promise<Object>} 测试结果
+   */
+  async triggerTest(accountId, platform, model = 'claude-sonnet-4-5-20250929') {
+    logger.info(`🎯 Manual test triggered for ${platform} account: ${accountId} (model: ${model})`)
+    return await this._runAccountTest(accountId, platform, model)
+  }
+
+  /**
+   * 获取账户测试历史
+   * @param {string} accountId - 账户ID
+   * @param {string} platform - 平台类型
+   * @returns {Promise<Array>} 测试历史
+   */
+  async getTestHistory(accountId, platform) {
+    return await redis.getAccountTestHistory(accountId, platform)
+  }
+
+  /**
+   * 获取账户测试配置
+   * @param {string} accountId - 账户ID
+   * @param {string} platform - 平台类型
+   * @returns {Promise<Object|null>}
+   */
+  async getTestConfig(accountId, platform) {
+    return await redis.getAccountTestConfig(accountId, platform)
+  }
+
+  /**
+   * 设置账户测试配置
+   * @param {string} accountId - 账户ID
+   * @param {string} platform - 平台类型
+   * @param {Object} testConfig - 测试配置 { enabled: boolean, cronExpression: string, model: string }
+   * @returns {Promise<void>}
+   */
+  async setTestConfig(accountId, platform, testConfig) {
+    // 验证 cron 表达式
+    if (testConfig.cronExpression && !this.validateCronExpression(testConfig.cronExpression)) {
+      throw new Error(`Invalid cron expression: ${testConfig.cronExpression}`)
+    }
+
+    await redis.saveAccountTestConfig(accountId, platform, testConfig)
+    logger.info(
+      `📝 Test config updated for ${platform} account ${accountId}: enabled=${testConfig.enabled}, cronExpression=${testConfig.cronExpression}, model=${testConfig.model}`
+    )
+
+    // 立即刷新任务，使配置立即生效
+    if (this.isStarted) {
+      await this._refreshAllTasks()
+    }
+  }
+
+  /**
+   * 更新单个账户的定时任务（配置变更时调用）
+   * @param {string} accountId
+   * @param {string} platform
+   */
+  async refreshAccountTask(accountId, platform) {
+    if (!this.isStarted) {
+      return
+    }
+
+    const accountKey = `${platform}:${accountId}`
+    const testConfig = await redis.getAccountTestConfig(accountId, platform)
+
+    // 停止现有任务
+    const existingTask = this.scheduledTasks.get(accountKey)
+    if (existingTask) {
+      existingTask.task.stop()
+      this.scheduledTasks.delete(accountKey)
+    }
+
+    // 如果启用且有有效的 cron 表达式，创建新任务
+    if (testConfig?.enabled && testConfig?.cronExpression) {
+      this._createCronTask(accountId, platform, testConfig.cronExpression, testConfig.model)
+      logger.info(
+        `🔄 Refreshed cron task for ${accountKey}: ${testConfig.cronExpression}, model: ${testConfig.model}`
+      )
+    }
+  }
+
+  /**
+   * 获取调度器状态
+   * @returns {Object}
+   */
+  getStatus() {
+    const tasks = []
+    for (const [accountKey, taskInfo] of this.scheduledTasks.entries()) {
+      tasks.push({
+        accountKey,
+        accountId: taskInfo.accountId,
+        platform: taskInfo.platform,
+        cronExpression: taskInfo.cronExpression,
+        model: taskInfo.model
+      })
+    }
+
+    return {
+      running: this.isStarted,
+      refreshIntervalMs: this.refreshIntervalMs,
+      scheduledTasksCount: this.scheduledTasks.size,
+      scheduledTasks: tasks,
+      currentlyTesting: Array.from(this.testingAccounts)
+    }
+  }
+}
+
+// 单例模式
+const accountTestSchedulerService = new AccountTestSchedulerService()
+
+module.exports = accountTestSchedulerService

src/services/bedrockRelayService.js

@@ -243,10 +243,11 @@ class BedrockRelayService {
             isBackendError ? { backendError: queueResult.errorMessage } : {}
           )
           if (!res.headersSent) {
+            const existingConnection = res.getHeader ? res.getHeader('Connection') : null
             res.writeHead(statusCode, {
               'Content-Type': 'text/event-stream',
               'Cache-Control': 'no-cache',
-              Connection: 'keep-alive',
+              Connection: existingConnection || 'keep-alive',
               'x-user-message-queue-error': errorType
             })
           }
@@ -309,10 +310,17 @@ class BedrockRelayService {
       }
 
       // 设置SSE响应头
+      // ⚠️ 关键修复：尊重 auth.js 提前设置的 Connection: close
+      const existingConnection = res.getHeader ? res.getHeader('Connection') : null
+      if (existingConnection) {
+        logger.debug(
+          `🔌 [Bedrock Stream] Preserving existing Connection header: ${existingConnection}`
+        )
+      }
       res.writeHead(200, {
         'Content-Type': 'text/event-stream',
         'Cache-Control': 'no-cache',
-        Connection: 'keep-alive',
+        Connection: existingConnection || 'keep-alive',
         'Access-Control-Allow-Origin': '*',
         'Access-Control-Allow-Headers': 'Content-Type, Authorization'
       })

src/services/ccrRelayService.js

@@ -4,6 +4,7 @@ const logger = require('../utils/logger')
 const config = require('../../config/config')
 const { parseVendorPrefixedModel } = require('../utils/modelHelper')
 const userMessageQueueService = require('./userMessageQueueService')
+const { isStreamWritable } = require('../utils/streamHelper')
 
 class CcrRelayService {
   constructor() {
@@ -379,10 +380,13 @@ class CcrRelayService {
             isBackendError ? { backendError: queueResult.errorMessage } : {}
           )
           if (!responseStream.headersSent) {
+            const existingConnection = responseStream.getHeader
+              ? responseStream.getHeader('Connection')
+              : null
             responseStream.writeHead(statusCode, {
               'Content-Type': 'text/event-stream',
               'Cache-Control': 'no-cache',
-              Connection: 'keep-alive',
+              Connection: existingConnection || 'keep-alive',
               'x-user-message-queue-error': errorType
             })
           }
@@ -606,10 +610,13 @@ class CcrRelayService {
 
             // 设置错误响应的状态码和响应头
             if (!responseStream.headersSent) {
+              const existingConnection = responseStream.getHeader
+                ? responseStream.getHeader('Connection')
+                : null
               const errorHeaders = {
                 'Content-Type': response.headers['content-type'] || 'application/json',
                 'Cache-Control': 'no-cache',
-                Connection: 'keep-alive'
+                Connection: existingConnection || 'keep-alive'
               }
               // 避免 Transfer-Encoding 冲突，让 Express 自动处理
               delete errorHeaders['Transfer-Encoding']
@@ -619,13 +626,13 @@ class CcrRelayService {
 
             // 直接透传错误数据，不进行包装
             response.data.on('data', (chunk) => {
-              if (!responseStream.destroyed) {
+              if (isStreamWritable(responseStream)) {
                 responseStream.write(chunk)
               }
             })
 
             response.data.on('end', () => {
-              if (!responseStream.destroyed) {
+              if (isStreamWritable(responseStream)) {
                 responseStream.end()
               }
               resolve() // 不抛出异常，正常完成流处理
@@ -659,11 +666,20 @@ class CcrRelayService {
           })
 
           // 设置响应头
+          // ⚠️ 关键修复：尊重 auth.js 提前设置的 Connection: close
           if (!responseStream.headersSent) {
+            const existingConnection = responseStream.getHeader
+              ? responseStream.getHeader('Connection')
+              : null
+            if (existingConnection) {
+              logger.debug(
+                `🔌 [CCR Stream] Preserving existing Connection header: ${existingConnection}`
+              )
+            }
             const headers = {
               'Content-Type': 'text/event-stream',
               'Cache-Control': 'no-cache',
-              Connection: 'keep-alive',
+              Connection: existingConnection || 'keep-alive',
               'Access-Control-Allow-Origin': '*',
               'Access-Control-Allow-Headers': 'Cache-Control'
             }
@@ -702,12 +718,17 @@ class CcrRelayService {
                   }
 
                   // 写入到响应流
-                  if (outputLine && !responseStream.destroyed) {
+                  if (outputLine && isStreamWritable(responseStream)) {
                     responseStream.write(`${outputLine}\n`)
+                  } else if (outputLine) {
+                    // 客户端连接已断开，记录警告
+                    logger.warn(
+                      `⚠️ [CCR] Client disconnected during stream, skipping data for account: ${accountId}`
+                    )
                   }
                 } else {
                   // 空行也需要传递
-                  if (!responseStream.destroyed) {
+                  if (isStreamWritable(responseStream)) {
                     responseStream.write('\n')
                   }
                 }
@@ -718,10 +739,6 @@ class CcrRelayService {
           })
 
           response.data.on('end', () => {
-            if (!responseStream.destroyed) {
-              responseStream.end()
-            }
-
             // 如果收集到使用统计数据，调用回调
             if (usageCallback && Object.keys(collectedUsage).length > 0) {
               try {
@@ -733,12 +750,26 @@ class CcrRelayService {
               }
             }
 
-            resolve()
+            if (isStreamWritable(responseStream)) {
+              // 等待数据完全 flush 到客户端后再 resolve
+              responseStream.end(() => {
+                logger.debug(
+                  `🌊 CCR stream response completed and flushed | bytesWritten: ${responseStream.bytesWritten || 'unknown'}`
+                )
+                resolve()
+              })
+            } else {
+              // 连接已断开，记录警告
+              logger.warn(
+                `⚠️ [CCR] Client disconnected before stream end, data may not have been received | account: ${accountId}`
+              )
+              resolve()
+            }
           })
 
           response.data.on('error', (err) => {
             logger.error('❌ Stream data error:', err)
-            if (!responseStream.destroyed) {
+            if (isStreamWritable(responseStream)) {
               responseStream.end()
             }
             reject(err)
@@ -770,7 +801,7 @@ class CcrRelayService {
             }
           }
 
-          if (!responseStream.destroyed) {
+          if (isStreamWritable(responseStream)) {
             responseStream.write(`data: ${JSON.stringify(errorResponse)}\n\n`)
             responseStream.end()
           }

src/services/claudeAccountService.js

@@ -91,7 +91,9 @@ class ClaudeAccountService {
       useUnifiedClientId = false, // 是否使用统一的客户端标识
       unifiedClientId = '', // 统一的客户端标识
       expiresAt = null, // 账户订阅到期时间
-      extInfo = null // 额外扩展信息
+      extInfo = null, // 额外扩展信息
+      maxConcurrency = 0, // 账户级用户消息串行队列：0=使用全局配置，>0=强制启用串行
+      interceptWarmup = false // 拦截预热请求（标题生成、Warmup等）
     } = options
 
     const accountId = uuidv4()
@@ -136,7 +138,11 @@ class ClaudeAccountService {
         // 账户订阅到期时间
         subscriptionExpiresAt: expiresAt || '',
         // 扩展信息
-        extInfo: normalizedExtInfo ? JSON.stringify(normalizedExtInfo) : ''
+        extInfo: normalizedExtInfo ? JSON.stringify(normalizedExtInfo) : '',
+        // 账户级用户消息串行队列限制
+        maxConcurrency: maxConcurrency.toString(),
+        // 拦截预热请求
+        interceptWarmup: interceptWarmup.toString()
       }
     } else {
       // 兼容旧格式
@@ -168,7 +174,11 @@ class ClaudeAccountService {
         // 账户订阅到期时间
         subscriptionExpiresAt: expiresAt || '',
         // 扩展信息
-        extInfo: normalizedExtInfo ? JSON.stringify(normalizedExtInfo) : ''
+        extInfo: normalizedExtInfo ? JSON.stringify(normalizedExtInfo) : '',
+        // 账户级用户消息串行队列限制
+        maxConcurrency: maxConcurrency.toString(),
+        // 拦截预热请求
+        interceptWarmup: interceptWarmup.toString()
       }
     }
 
@@ -216,7 +226,8 @@ class ClaudeAccountService {
       useUnifiedUserAgent,
       useUnifiedClientId,
       unifiedClientId,
-      extInfo: normalizedExtInfo
+      extInfo: normalizedExtInfo,
+      interceptWarmup
     }
   }
 
@@ -574,7 +585,11 @@ class ClaudeAccountService {
             // 添加停止原因
             stoppedReason: account.stoppedReason || null,
             // 扩展信息
-            extInfo: parsedExtInfo
+            extInfo: parsedExtInfo,
+            // 账户级用户消息串行队列限制
+            maxConcurrency: parseInt(account.maxConcurrency || '0', 10),
+            // 拦截预热请求
+            interceptWarmup: account.interceptWarmup === 'true'
           }
         })
       )
@@ -666,7 +681,9 @@ class ClaudeAccountService {
         'useUnifiedClientId',
         'unifiedClientId',
         'subscriptionExpiresAt',
-        'extInfo'
+        'extInfo',
+        'maxConcurrency',
+        'interceptWarmup'
       ]
       const updatedData = { ...accountData }
       let shouldClearAutoStopFields = false
@@ -681,7 +698,7 @@ class ClaudeAccountService {
             updatedData[field] = this._encryptSensitiveData(value)
           } else if (field === 'proxy') {
             updatedData[field] = value ? JSON.stringify(value) : ''
-          } else if (field === 'priority') {
+          } else if (field === 'priority' || field === 'maxConcurrency') {
             updatedData[field] = value.toString()
           } else if (field === 'subscriptionInfo') {
             // 处理订阅信息更新

src/services/claudeConsoleAccountService.js

@@ -68,7 +68,8 @@ class ClaudeConsoleAccountService {
       dailyQuota = 0, // 每日额度限制（美元），0表示不限制
       quotaResetTime = '00:00', // 额度重置时间（HH:mm格式）
       maxConcurrentTasks = 0, // 最大并发任务数，0表示无限制
-      disableAutoProtection = false // 是否关闭自动防护（429/401/400/529 不自动禁用）
+      disableAutoProtection = false, // 是否关闭自动防护（429/401/400/529 不自动禁用）
+      interceptWarmup = false // 拦截预热请求（标题生成、Warmup等）
     } = options
 
     // 验证必填字段
@@ -117,7 +118,8 @@ class ClaudeConsoleAccountService {
       quotaResetTime, // 额度重置时间
       quotaStoppedAt: '', // 因额度停用的时间
       maxConcurrentTasks: maxConcurrentTasks.toString(), // 最大并发任务数，0表示无限制
-      disableAutoProtection: disableAutoProtection.toString() // 关闭自动防护
+      disableAutoProtection: disableAutoProtection.toString(), // 关闭自动防护
+      interceptWarmup: interceptWarmup.toString() // 拦截预热请求
     }
 
     const client = redis.getClientSafe()
@@ -156,6 +158,7 @@ class ClaudeConsoleAccountService {
       quotaStoppedAt: null,
       maxConcurrentTasks, // 新增：返回并发限制配置
       disableAutoProtection, // 新增：返回自动防护开关
+      interceptWarmup, // 新增：返回预热请求拦截开关
       activeTaskCount: 0 // 新增：新建账户当前并发数为0
     }
   }
@@ -217,7 +220,9 @@ class ClaudeConsoleAccountService {
             // 并发控制相关
             maxConcurrentTasks: parseInt(accountData.maxConcurrentTasks) || 0,
             activeTaskCount,
-            disableAutoProtection: accountData.disableAutoProtection === 'true'
+            disableAutoProtection: accountData.disableAutoProtection === 'true',
+            // 拦截预热请求
+            interceptWarmup: accountData.interceptWarmup === 'true'
           })
         }
       }
@@ -375,6 +380,9 @@ class ClaudeConsoleAccountService {
       if (updates.disableAutoProtection !== undefined) {
         updatedData.disableAutoProtection = updates.disableAutoProtection.toString()
       }
+      if (updates.interceptWarmup !== undefined) {
+        updatedData.interceptWarmup = updates.interceptWarmup.toString()
+      }
 
       // ✅ 直接保存 subscriptionExpiresAt（如果提供）
       // Claude Console 没有 token 刷新逻辑，不会覆盖此字段

src/services/claudeConsoleRelayService.js

@@ -10,6 +10,8 @@ const {
   isAccountDisabledError
 } = require('../utils/errorSanitizer')
 const userMessageQueueService = require('./userMessageQueueService')
+const { isStreamWritable } = require('../utils/streamHelper')
+const { filterForClaude } = require('../utils/headerFilter')
 
 class ClaudeConsoleRelayService {
   constructor() {
@@ -517,10 +519,13 @@ class ClaudeConsoleRelayService {
             isBackendError ? { backendError: queueResult.errorMessage } : {}
           )
           if (!responseStream.headersSent) {
+            const existingConnection = responseStream.getHeader
+              ? responseStream.getHeader('Connection')
+              : null
             responseStream.writeHead(statusCode, {
               'Content-Type': 'text/event-stream',
               'Cache-Control': 'no-cache',
-              Connection: 'keep-alive',
+              Connection: existingConnection || 'keep-alive',
               'x-user-message-queue-error': errorType
             })
           }
@@ -878,7 +883,7 @@ class ClaudeConsoleRelayService {
                   `🧹 [Stream] [SANITIZED] Error response to client: ${JSON.stringify(sanitizedError)}`
                 )
 
-                if (!responseStream.destroyed) {
+                if (isStreamWritable(responseStream)) {
                   responseStream.write(JSON.stringify(sanitizedError))
                   responseStream.end()
                 }
@@ -886,7 +891,7 @@ class ClaudeConsoleRelayService {
                 const sanitizedText = sanitizeErrorMessage(errorDataForCheck)
                 logger.error(`🧹 [Stream] [SANITIZED] Error response to client: ${sanitizedText}`)
 
-                if (!responseStream.destroyed) {
+                if (isStreamWritable(responseStream)) {
                   responseStream.write(sanitizedText)
                   responseStream.end()
                 }
@@ -923,11 +928,22 @@ class ClaudeConsoleRelayService {
           })
 
           // 设置响应头
+          // ⚠️ 关键修复：尊重 auth.js 提前设置的 Connection: close
+          // 当并发队列功能启用时，auth.js 会设置 Connection: close 来禁用 Keep-Alive
           if (!responseStream.headersSent) {
+            const existingConnection = responseStream.getHeader
+              ? responseStream.getHeader('Connection')
+              : null
+            const connectionHeader = existingConnection || 'keep-alive'
+            if (existingConnection) {
+              logger.debug(
+                `🔌 [Console Stream] Preserving existing Connection header: ${existingConnection}`
+              )
+            }
             responseStream.writeHead(200, {
               'Content-Type': 'text/event-stream',
               'Cache-Control': 'no-cache',
-              Connection: 'keep-alive',
+              Connection: connectionHeader,
               'X-Accel-Buffering': 'no'
             })
           }
@@ -953,20 +969,33 @@ class ClaudeConsoleRelayService {
               buffer = lines.pop() || ''
 
               // 转发数据并解析usage
-              if (lines.length > 0 && !responseStream.destroyed) {
-                const linesToForward = lines.join('\n') + (lines.length > 0 ? '\n' : '')
+              if (lines.length > 0) {
+                // 检查流是否可写（客户端连接是否有效）
+                if (isStreamWritable(responseStream)) {
+                  const linesToForward = lines.join('\n') + (lines.length > 0 ? '\n' : '')
 
-                // 应用流转换器如果有
-                if (streamTransformer) {
-                  const transformed = streamTransformer(linesToForward)
-                  if (transformed) {
-                    responseStream.write(transformed)
+                  // 应用流转换器如果有
+                  let dataToWrite = linesToForward
+                  if (streamTransformer) {
+                    const transformed = streamTransformer(linesToForward)
+                    if (transformed) {
+                      dataToWrite = transformed
+                    } else {
+                      dataToWrite = null
+                    }
+                  }
+
+                  if (dataToWrite) {
+                    responseStream.write(dataToWrite)
                   }
                 } else {
-                  responseStream.write(linesToForward)
+                  // 客户端连接已断开，记录警告（但仍继续解析usage）
+                  logger.warn(
+                    `⚠️ [Console] Client disconnected during stream, skipping ${lines.length} lines for account: ${account?.name || accountId}`
+                  )
                 }
 
-                // 解析SSE数据寻找usage信息
+                // 解析SSE数据寻找usage信息（无论连接状态如何）
                 for (const line of lines) {
                   if (line.startsWith('data:')) {
                     const jsonStr = line.slice(5).trimStart()
@@ -1074,7 +1103,7 @@ class ClaudeConsoleRelayService {
                 `❌ Error processing Claude Console stream data (Account: ${account?.name || accountId}):`,
                 error
               )
-              if (!responseStream.destroyed) {
+              if (isStreamWritable(responseStream)) {
                 // 如果有 streamTransformer（如测试请求），使用前端期望的格式
                 if (streamTransformer) {
                   responseStream.write(
@@ -1097,7 +1126,7 @@ class ClaudeConsoleRelayService {
           response.data.on('end', () => {
             try {
               // 处理缓冲区中剩余的数据
-              if (buffer.trim() && !responseStream.destroyed) {
+              if (buffer.trim() && isStreamWritable(responseStream)) {
                 if (streamTransformer) {
                   const transformed = streamTransformer(buffer)
                   if (transformed) {
@@ -1146,12 +1175,33 @@ class ClaudeConsoleRelayService {
               }
 
               // 确保流正确结束
-              if (!responseStream.destroyed) {
-                responseStream.end()
-              }
+              if (isStreamWritable(responseStream)) {
+                // 📊 诊断日志：流结束前状态
+                logger.info(
+                  `📤 [STREAM] Ending response | destroyed: ${responseStream.destroyed}, ` +
+                    `socketDestroyed: ${responseStream.socket?.destroyed}, ` +
+                    `socketBytesWritten: ${responseStream.socket?.bytesWritten || 0}`
+                )
 
-              logger.debug('🌊 Claude Console Claude stream response completed')
-              resolve()
+                // 禁用 Nagle 算法确保数据立即发送
+                if (responseStream.socket && !responseStream.socket.destroyed) {
+                  responseStream.socket.setNoDelay(true)
+                }
+
+                // 等待数据完全 flush 到客户端后再 resolve
+                responseStream.end(() => {
+                  logger.info(
+                    `✅ [STREAM] Response ended and flushed | socketBytesWritten: ${responseStream.socket?.bytesWritten || 'unknown'}`
+                  )
+                  resolve()
+                })
+              } else {
+                // 连接已断开，记录警告
+                logger.warn(
+                  `⚠️ [Console] Client disconnected before stream end, data may not have been received | account: ${account?.name || accountId}`
+                )
+                resolve()
+              }
             } catch (error) {
               logger.error('❌ Error processing stream end:', error)
               reject(error)
@@ -1163,7 +1213,7 @@ class ClaudeConsoleRelayService {
               `❌ Claude Console stream error (Account: ${account?.name || accountId}):`,
               error
             )
-            if (!responseStream.destroyed) {
+            if (isStreamWritable(responseStream)) {
               // 如果有 streamTransformer（如测试请求），使用前端期望的格式
               if (streamTransformer) {
                 responseStream.write(
@@ -1211,14 +1261,17 @@ class ClaudeConsoleRelayService {
 
           // 发送错误响应
           if (!responseStream.headersSent) {
+            const existingConnection = responseStream.getHeader
+              ? responseStream.getHeader('Connection')
+              : null
             responseStream.writeHead(error.response?.status || 500, {
               'Content-Type': 'text/event-stream',
               'Cache-Control': 'no-cache',
-              Connection: 'keep-alive'
+              Connection: existingConnection || 'keep-alive'
             })
           }
 
-          if (!responseStream.destroyed) {
+          if (isStreamWritable(responseStream)) {
             // 如果有 streamTransformer（如测试请求），使用前端期望的格式
             if (streamTransformer) {
               responseStream.write(
@@ -1250,30 +1303,9 @@ class ClaudeConsoleRelayService {
 
   // 🔧 过滤客户端请求头
   _filterClientHeaders(clientHeaders) {
-    const sensitiveHeaders = [
-      'content-type',
-      'user-agent',
-      'authorization',
-      'x-api-key',
-      'host',
-      'content-length',
-      'connection',
-      'proxy-authorization',
-      'content-encoding',
-      'transfer-encoding',
-      'anthropic-version'
-    ]
-
-    const filteredHeaders = {}
-
-    Object.keys(clientHeaders || {}).forEach((key) => {
-      const lowerKey = key.toLowerCase()
-      if (!sensitiveHeaders.includes(lowerKey)) {
-        filteredHeaders[key] = clientHeaders[key]
-      }
-    })
-
-    return filteredHeaders
+    // 使用统一的 headerFilter 工具类（白名单模式）
+    // 与 claudeRelayService 保持一致，避免透传 CDN headers 触发上游 API 安全检查
+    return filterForClaude(clientHeaders)
   }
 
   // 🕐 更新最后使用时间
@@ -1388,7 +1420,7 @@ class ClaudeConsoleRelayService {
           'Cache-Control': 'no-cache'
         })
       }
-      if (!responseStream.destroyed && !responseStream.writableEnded) {
+      if (isStreamWritable(responseStream)) {
         responseStream.write(
           `data: ${JSON.stringify({ type: 'test_complete', success: false, error: error.message })}\n\n`
         )

src/services/claudeRelayConfigService.js

@@ -20,6 +20,15 @@ const DEFAULT_CONFIG = {
   userMessageQueueDelayMs: 200, // 请求间隔（毫秒）
   userMessageQueueTimeoutMs: 5000, // 队列等待超时（毫秒），优化后锁持有时间短无需长等待
   userMessageQueueLockTtlMs: 5000, // 锁TTL（毫秒），请求发送后立即释放无需长TTL
+  // 并发请求排队配置
+  concurrentRequestQueueEnabled: false, // 是否启用并发请求排队（默认关闭）
+  concurrentRequestQueueMaxSize: 3, // 固定最小排队数（默认3）
+  concurrentRequestQueueMaxSizeMultiplier: 0, // 并发数的倍数（默认0，仅使用固定值）
+  concurrentRequestQueueTimeoutMs: 10000, // 排队超时（毫秒，默认10秒）
+  concurrentRequestQueueMaxRedisFailCount: 5, // 连续 Redis 失败阈值（默认5次）
+  // 排队健康检查配置
+  concurrentRequestQueueHealthCheckEnabled: true, // 是否启用排队健康检查（默认开启）
+  concurrentRequestQueueHealthThreshold: 0.8, // 健康检查阈值（P90 >= 超时 × 阈值时拒绝新请求）
   updatedAt: null,
   updatedBy: null
 }
@@ -105,7 +114,8 @@ class ClaudeRelayConfigService {
 
       logger.info(`✅ Claude relay config updated by ${updatedBy}:`, {
         claudeCodeOnlyEnabled: updatedConfig.claudeCodeOnlyEnabled,
-        globalSessionBindingEnabled: updatedConfig.globalSessionBindingEnabled
+        globalSessionBindingEnabled: updatedConfig.globalSessionBindingEnabled,
+        concurrentRequestQueueEnabled: updatedConfig.concurrentRequestQueueEnabled
       })
 
       return updatedConfig

src/services/claudeRelayService.js

@@ -16,6 +16,7 @@ const { formatDateWithTimezone } = require('../utils/dateHelper')
 const requestIdentityService = require('./requestIdentityService')
 const { createClaudeTestPayload } = require('../utils/testPayloadHelper')
 const userMessageQueueService = require('./userMessageQueueService')
+const { isStreamWritable } = require('../utils/streamHelper')
 
 class ClaudeRelayService {
   constructor() {
@@ -209,7 +210,17 @@ class ClaudeRelayService {
           logger.error('❌ accountId missing for queue lock in relayRequest')
           throw new Error('accountId missing for queue lock')
         }
-        const queueResult = await userMessageQueueService.acquireQueueLock(accountId)
+        // 获取账户信息以检查账户级串行队列配置
+        const accountForQueue = await claudeAccountService.getAccount(accountId)
+        const accountConfig = accountForQueue
+          ? { maxConcurrency: parseInt(accountForQueue.maxConcurrency || '0', 10) }
+          : null
+        const queueResult = await userMessageQueueService.acquireQueueLock(
+          accountId,
+          null,
+          null,
+          accountConfig
+        )
         if (!queueResult.acquired && !queueResult.skipped) {
           // 区分 Redis 后端错误和队列超时
           const isBackendError = queueResult.error === 'queue_backend_error'
@@ -322,17 +333,46 @@ class ClaudeRelayService {
       }
 
       // 发送请求到Claude API（传入回调以获取请求对象）
-      const response = await this._makeClaudeRequest(
-        processedBody,
-        accessToken,
-        proxyAgent,
-        clientHeaders,
-        accountId,
-        (req) => {
-          upstreamRequest = req
-        },
-        options
-      )
+      // 🔄 403 重试机制：仅对 claude-official 类型账户（OAuth 或 Setup Token）
+      const maxRetries = this._shouldRetryOn403(accountType) ? 2 : 0
+      let retryCount = 0
+      let response
+      let shouldRetry = false
+
+      do {
+        response = await this._makeClaudeRequest(
+          processedBody,
+          accessToken,
+          proxyAgent,
+          clientHeaders,
+          accountId,
+          (req) => {
+            upstreamRequest = req
+          },
+          options
+        )
+
+        // 检查是否需要重试 403
+        shouldRetry = response.statusCode === 403 && retryCount < maxRetries
+        if (shouldRetry) {
+          retryCount++
+          logger.warn(
+            `🔄 403 error for account ${accountId}, retry ${retryCount}/${maxRetries} after 2s`
+          )
+          await this._sleep(2000)
+        }
+      } while (shouldRetry)
+
+      // 如果进行了重试，记录最终结果
+      if (retryCount > 0) {
+        if (response.statusCode === 403) {
+          logger.error(`🚫 403 error persists for account ${accountId} after ${retryCount} retries`)
+        } else {
+          logger.info(
+            `✅ 403 retry successful for account ${accountId} on attempt ${retryCount}, got status ${response.statusCode}`
+          )
+        }
+      }
 
       // 📬 请求已发送成功，立即释放队列锁（无需等待响应处理完成）
       // 因为 Claude API 限流基于请求发送时刻计算（RPM），不是请求完成时刻
@@ -397,9 +437,10 @@ class ClaudeRelayService {
           }
         }
         // 检查是否为403状态码（禁止访问）
+        // 注意：如果进行了重试，retryCount > 0；这里的 403 是重试后最终的结果
         else if (response.statusCode === 403) {
           logger.error(
-            `🚫 Forbidden error (403) detected for account ${accountId}, marking as blocked`
+            `🚫 Forbidden error (403) detected for account ${accountId}${retryCount > 0 ? ` after ${retryCount} retries` : ''}, marking as blocked`
           )
           await unifiedClaudeScheduler.markAccountBlocked(accountId, accountType, sessionHash)
         }
@@ -1057,6 +1098,8 @@ class ClaudeRelayService {
 
     logger.info(`🔗 指纹是这个: ${headers['User-Agent']}`)
 
+    logger.info(`🔗 指纹是这个: ${headers['User-Agent']}`)
+
     // 根据模型和客户端传递的 anthropic-beta 动态设置 header
     const modelId = requestPayload?.model || body?.model
     const clientBetaHeader = clientHeaders?.['anthropic-beta']
@@ -1311,7 +1354,17 @@ class ClaudeRelayService {
           logger.error('❌ accountId missing for queue lock in relayStreamRequestWithUsageCapture')
           throw new Error('accountId missing for queue lock')
         }
-        const queueResult = await userMessageQueueService.acquireQueueLock(accountId)
+        // 获取账户信息以检查账户级串行队列配置
+        const accountForQueue = await claudeAccountService.getAccount(accountId)
+        const accountConfig = accountForQueue
+          ? { maxConcurrency: parseInt(accountForQueue.maxConcurrency || '0', 10) }
+          : null
+        const queueResult = await userMessageQueueService.acquireQueueLock(
+          accountId,
+          null,
+          null,
+          accountConfig
+        )
         if (!queueResult.acquired && !queueResult.skipped) {
           // 区分 Redis 后端错误和队列超时
           const isBackendError = queueResult.error === 'queue_backend_error'
@@ -1338,10 +1391,13 @@ class ClaudeRelayService {
             isBackendError ? { backendError: queueResult.errorMessage } : {}
           )
           if (!responseStream.headersSent) {
+            const existingConnection = responseStream.getHeader
+              ? responseStream.getHeader('Connection')
+              : null
             responseStream.writeHead(statusCode, {
               'Content-Type': 'text/event-stream',
               'Cache-Control': 'no-cache',
-              Connection: 'keep-alive',
+              Connection: existingConnection || 'keep-alive',
               'x-user-message-queue-error': errorType
             })
           }
@@ -1491,8 +1547,10 @@ class ClaudeRelayService {
     streamTransformer = null,
     requestOptions = {},
     isDedicatedOfficialAccount = false,
-    onResponseStart = null // 📬 新增：收到响应头时的回调，用于提前释放队列锁
+    onResponseStart = null, // 📬 新增：收到响应头时的回调，用于提前释放队列锁
+    retryCount = 0 // 🔄 403 重试计数器
   ) {
+    const maxRetries = 2 // 最大重试次数
     // 获取账户信息用于统一 User-Agent
     const account = await claudeAccountService.getAccount(accountId)
 
@@ -1605,6 +1663,51 @@ class ClaudeRelayService {
             }
           }
 
+          // 🔄 403 重试机制（必须在设置 res.on('data')/res.on('end') 之前处理）
+          // 否则重试时旧响应的 on('end') 会与新请求产生竞态条件
+          if (res.statusCode === 403) {
+            const canRetry =
+              this._shouldRetryOn403(accountType) &&
+              retryCount < maxRetries &&
+              !responseStream.headersSent
+
+            if (canRetry) {
+              logger.warn(
+                `🔄 [Stream] 403 error for account ${accountId}, retry ${retryCount + 1}/${maxRetries} after 2s`
+              )
+              // 消费当前响应并销毁请求
+              res.resume()
+              req.destroy()
+
+              // 等待 2 秒后递归重试
+              await this._sleep(2000)
+
+              try {
+                // 递归调用自身进行重试
+                const retryResult = await this._makeClaudeStreamRequestWithUsageCapture(
+                  body,
+                  accessToken,
+                  proxyAgent,
+                  clientHeaders,
+                  responseStream,
+                  usageCallback,
+                  accountId,
+                  accountType,
+                  sessionHash,
+                  streamTransformer,
+                  requestOptions,
+                  isDedicatedOfficialAccount,
+                  onResponseStart,
+                  retryCount + 1
+                )
+                resolve(retryResult)
+              } catch (retryError) {
+                reject(retryError)
+              }
+              return // 重要：提前返回，不设置后续的错误处理器
+            }
+          }
+
           // 将错误处理逻辑封装在一个异步函数中
           const handleErrorResponse = async () => {
             if (res.statusCode === 401) {
@@ -1628,8 +1731,10 @@ class ClaudeRelayService {
                 )
               }
             } else if (res.statusCode === 403) {
+              // 403 处理：走到这里说明重试已用尽或不适用重试，直接标记 blocked
+              // 注意：重试逻辑已在 handleErrorResponse 外部提前处理
               logger.error(
-                `🚫 [Stream] Forbidden error (403) detected for account ${accountId}, marking as blocked`
+                `🚫 [Stream] Forbidden error (403) detected for account ${accountId}${retryCount > 0 ? ` after ${retryCount} retries` : ''}, marking as blocked`
               )
               await unifiedClaudeScheduler.markAccountBlocked(accountId, accountType, sessionHash)
             } else if (res.statusCode === 529) {
@@ -1699,7 +1804,7 @@ class ClaudeRelayService {
                 }
               })()
             }
-            if (!responseStream.destroyed) {
+            if (isStreamWritable(responseStream)) {
               // 解析 Claude API 返回的错误详情
               let errorMessage = `Claude API error: ${res.statusCode}`
               try {
@@ -1764,16 +1869,23 @@ class ClaudeRelayService {
             buffer = lines.pop() || '' // 保留最后的不完整行
 
             // 转发已处理的完整行到客户端
-            if (lines.length > 0 && !responseStream.destroyed) {
-              const linesToForward = lines.join('\n') + (lines.length > 0 ? '\n' : '')
-              // 如果有流转换器，应用转换
-              if (streamTransformer) {
-                const transformed = streamTransformer(linesToForward)
-                if (transformed) {
-                  responseStream.write(transformed)
+            if (lines.length > 0) {
+              if (isStreamWritable(responseStream)) {
+                const linesToForward = lines.join('\n') + (lines.length > 0 ? '\n' : '')
+                // 如果有流转换器，应用转换
+                if (streamTransformer) {
+                  const transformed = streamTransformer(linesToForward)
+                  if (transformed) {
+                    responseStream.write(transformed)
+                  }
+                } else {
+                  responseStream.write(linesToForward)
                 }
               } else {
-                responseStream.write(linesToForward)
+                // 客户端连接已断开，记录警告（但仍继续解析usage）
+                logger.warn(
+                  `⚠️ [Official] Client disconnected during stream, skipping ${lines.length} lines for account: ${accountId}`
+                )
               }
             }
 
@@ -1878,7 +1990,7 @@ class ClaudeRelayService {
           } catch (error) {
             logger.error('❌ Error processing stream data:', error)
             // 发送错误但不破坏流，让它自然结束
-            if (!responseStream.destroyed) {
+            if (isStreamWritable(responseStream)) {
               responseStream.write('event: error\n')
               responseStream.write(
                 `data: ${JSON.stringify({
@@ -1894,7 +2006,7 @@ class ClaudeRelayService {
         res.on('end', async () => {
           try {
             // 处理缓冲区中剩余的数据
-            if (buffer.trim() && !responseStream.destroyed) {
+            if (buffer.trim() && isStreamWritable(responseStream)) {
               if (streamTransformer) {
                 const transformed = streamTransformer(buffer)
                 if (transformed) {
@@ -1906,8 +2018,16 @@ class ClaudeRelayService {
             }
 
             // 确保流正确结束
-            if (!responseStream.destroyed) {
+            if (isStreamWritable(responseStream)) {
               responseStream.end()
+              logger.debug(
+                `🌊 Stream end called | bytesWritten: ${responseStream.bytesWritten || 'unknown'}`
+              )
+            } else {
+              // 连接已断开，记录警告
+              logger.warn(
+                `⚠️ [Official] Client disconnected before stream end, data may not have been received | account: ${account?.name || accountId}`
+              )
             }
           } catch (error) {
             logger.error('❌ Error processing stream end:', error)
@@ -2105,14 +2225,17 @@ class ClaudeRelayService {
         }
 
         if (!responseStream.headersSent) {
+          const existingConnection = responseStream.getHeader
+            ? responseStream.getHeader('Connection')
+            : null
           responseStream.writeHead(statusCode, {
             'Content-Type': 'text/event-stream',
             'Cache-Control': 'no-cache',
-            Connection: 'keep-alive'
+            Connection: existingConnection || 'keep-alive'
           })
         }
 
-        if (!responseStream.destroyed) {
+        if (isStreamWritable(responseStream)) {
           // 发送 SSE 错误事件
           responseStream.write('event: error\n')
           responseStream.write(
@@ -2132,13 +2255,16 @@ class ClaudeRelayService {
         logger.error(`❌ Claude stream request timeout | Account: ${account?.name || accountId}`)
 
         if (!responseStream.headersSent) {
+          const existingConnection = responseStream.getHeader
+            ? responseStream.getHeader('Connection')
+            : null
           responseStream.writeHead(504, {
             'Content-Type': 'text/event-stream',
             'Cache-Control': 'no-cache',
-            Connection: 'keep-alive'
+            Connection: existingConnection || 'keep-alive'
           })
         }
-        if (!responseStream.destroyed) {
+        if (isStreamWritable(responseStream)) {
           // 发送 SSE 错误事件
           responseStream.write('event: error\n')
           responseStream.write(
@@ -2429,34 +2555,44 @@ class ClaudeRelayService {
     }
   }
 
+  // 🔧 准备测试请求的公共逻辑（供 testAccountConnection 和 testAccountConnectionSync 共用）
+  async _prepareAccountForTest(accountId) {
+    // 获取账户信息
+    const account = await claudeAccountService.getAccount(accountId)
+    if (!account) {
+      throw new Error('Account not found')
+    }
+
+    // 获取有效的访问token
+    const accessToken = await claudeAccountService.getValidAccessToken(accountId)
+    if (!accessToken) {
+      throw new Error('Failed to get valid access token')
+    }
+
+    // 获取代理配置
+    const proxyAgent = await this._getProxyAgent(accountId)
+
+    return { account, accessToken, proxyAgent }
+  }
+
   // 🧪 测试账号连接（供Admin API使用，直接复用 _makeClaudeStreamRequestWithUsageCapture）
-  async testAccountConnection(accountId, responseStream) {
-    const testRequestBody = createClaudeTestPayload('claude-sonnet-4-5-20250929', { stream: true })
+  async testAccountConnection(accountId, responseStream, model = 'claude-sonnet-4-5-20250929') {
+    const testRequestBody = createClaudeTestPayload(model, { stream: true })
 
     try {
-      // 获取账户信息
-      const account = await claudeAccountService.getAccount(accountId)
-      if (!account) {
-        throw new Error('Account not found')
-      }
+      const { account, accessToken, proxyAgent } = await this._prepareAccountForTest(accountId)
 
       logger.info(`🧪 Testing Claude account connection: ${account.name} (${accountId})`)
 
-      // 获取有效的访问token
-      const accessToken = await claudeAccountService.getValidAccessToken(accountId)
-      if (!accessToken) {
-        throw new Error('Failed to get valid access token')
-      }
-
-      // 获取代理配置
-      const proxyAgent = await this._getProxyAgent(accountId)
-
       // 设置响应头
       if (!responseStream.headersSent) {
+        const existingConnection = responseStream.getHeader
+          ? responseStream.getHeader('Connection')
+          : null
         responseStream.writeHead(200, {
           'Content-Type': 'text/event-stream',
           'Cache-Control': 'no-cache',
-          Connection: 'keep-alive',
+          Connection: existingConnection || 'keep-alive',
           'X-Accel-Buffering': 'no'
         })
       }
@@ -2484,7 +2620,7 @@ class ClaudeRelayService {
     } catch (error) {
       logger.error(`❌ Test account connection failed:`, error)
       // 发送错误事件给前端
-      if (!responseStream.destroyed && !responseStream.writableEnded) {
+      if (isStreamWritable(responseStream)) {
         try {
           const errorMsg = error.message || '测试失败'
           responseStream.write(`data: ${JSON.stringify({ type: 'error', error: errorMsg })}\n\n`)
@@ -2496,6 +2632,125 @@ class ClaudeRelayService {
     }
   }
 
+  // 🧪 非流式测试账号连接（供定时任务使用）
+  // 复用流式请求方法，收集结果后返回
+  async testAccountConnectionSync(accountId, model = 'claude-sonnet-4-5-20250929') {
+    const testRequestBody = createClaudeTestPayload(model, { stream: true })
+    const startTime = Date.now()
+
+    try {
+      // 使用公共方法准备测试所需的账户信息、token 和代理
+      const { account, accessToken, proxyAgent } = await this._prepareAccountForTest(accountId)
+
+      logger.info(`🧪 Testing Claude account connection (sync): ${account.name} (${accountId})`)
+
+      // 创建一个收集器来捕获流式响应
+      let responseText = ''
+      let capturedUsage = null
+      let capturedModel = model
+      let hasError = false
+      let errorMessage = ''
+
+      // 创建模拟的响应流对象
+      const mockResponseStream = {
+        headersSent: true, // 跳过设置响应头
+        write: (data) => {
+          // 解析 SSE 数据
+          if (typeof data === 'string' && data.startsWith('data: ')) {
+            try {
+              const jsonStr = data.replace('data: ', '').trim()
+              if (jsonStr && jsonStr !== '[DONE]') {
+                const parsed = JSON.parse(jsonStr)
+                // 提取文本内容
+                if (parsed.type === 'content_block_delta' && parsed.delta?.text) {
+                  responseText += parsed.delta.text
+                }
+                // 提取 usage 信息
+                if (parsed.type === 'message_delta' && parsed.usage) {
+                  capturedUsage = parsed.usage
+                }
+                // 提取模型信息
+                if (parsed.type === 'message_start' && parsed.message?.model) {
+                  capturedModel = parsed.message.model
+                }
+                // 检测错误
+                if (parsed.type === 'error') {
+                  hasError = true
+                  errorMessage = parsed.error?.message || 'Unknown error'
+                }
+              }
+            } catch {
+              // 忽略解析错误
+            }
+          }
+          return true
+        },
+        end: () => {},
+        on: () => {},
+        once: () => {},
+        emit: () => {},
+        writable: true
+      }
+
+      // 复用流式请求方法
+      await this._makeClaudeStreamRequestWithUsageCapture(
+        testRequestBody,
+        accessToken,
+        proxyAgent,
+        {}, // clientHeaders - 测试不需要客户端headers
+        mockResponseStream,
+        null, // usageCallback - 测试不需要统计
+        accountId,
+        'claude-official', // accountType
+        null, // sessionHash - 测试不需要会话
+        null, // streamTransformer - 不需要转换，直接解析原始格式
+        {}, // requestOptions
+        false // isDedicatedOfficialAccount
+      )
+
+      const latencyMs = Date.now() - startTime
+
+      if (hasError) {
+        logger.warn(`⚠️ Test completed with error for account: ${account.name} - ${errorMessage}`)
+        return {
+          success: false,
+          error: errorMessage,
+          latencyMs,
+          timestamp: new Date().toISOString()
+        }
+      }
+
+      logger.info(`✅ Test completed for account: ${account.name} (${latencyMs}ms)`)
+
+      return {
+        success: true,
+        message: responseText.substring(0, 200), // 截取前200字符
+        latencyMs,
+        model: capturedModel,
+        usage: capturedUsage,
+        timestamp: new Date().toISOString()
+      }
+    } catch (error) {
+      const latencyMs = Date.now() - startTime
+      logger.error(`❌ Test account connection (sync) failed:`, error.message)
+
+      // 提取错误详情
+      let errorMessage = error.message
+      if (error.response) {
+        errorMessage =
+          error.response.data?.error?.message || error.response.statusText || error.message
+      }
+
+      return {
+        success: false,
+        error: errorMessage,
+        statusCode: error.response?.status,
+        latencyMs,
+        timestamp: new Date().toISOString()
+      }
+    }
+  }
+
   // 🎯 健康检查
   async healthCheck() {
     try {
@@ -2517,6 +2772,17 @@ class ClaudeRelayService {
       }
     }
   }
+
+  // 🔄 判断账户是否应该在 403 错误时进行重试
+  // 仅 claude-official 类型账户（OAuth 或 Setup Token 授权）需要重试
+  _shouldRetryOn403(accountType) {
+    return accountType === 'claude-official'
+  }
+
+  // ⏱️ 等待指定毫秒数
+  _sleep(ms) {
+    return new Promise((resolve) => setTimeout(resolve, ms))
+  }
 }
 
 module.exports = new ClaudeRelayService()

src/services/unifiedOpenAIScheduler.js

@@ -9,6 +9,26 @@ class UnifiedOpenAIScheduler {
     this.SESSION_MAPPING_PREFIX = 'unified_openai_session_mapping:'
   }
 
+  // 🔢 按优先级和最后使用时间排序账户（与 Claude/Gemini 调度保持一致）
+  _sortAccountsByPriority(accounts) {
+    return accounts.sort((a, b) => {
+      const aPriority = Number.parseInt(a.priority, 10)
+      const bPriority = Number.parseInt(b.priority, 10)
+      const normalizedAPriority = Number.isFinite(aPriority) ? aPriority : 50
+      const normalizedBPriority = Number.isFinite(bPriority) ? bPriority : 50
+
+      // 首先按优先级排序（数字越小优先级越高）
+      if (normalizedAPriority !== normalizedBPriority) {
+        return normalizedAPriority - normalizedBPriority
+      }
+
+      // 优先级相同时，按最后使用时间排序（最久未使用的优先）
+      const aLastUsed = new Date(a.lastUsedAt || 0).getTime()
+      const bLastUsed = new Date(b.lastUsedAt || 0).getTime()
+      return aLastUsed - bLastUsed
+    })
+  }
+
   // 🔧 辅助方法：检查账户是否可调度（兼容字符串和布尔值）
   _isSchedulable(schedulable) {
     // 如果是 undefined 或 null，默认为可调度
@@ -244,13 +264,7 @@ class UnifiedOpenAIScheduler {
             `🎯 Using bound dedicated ${accountType} account: ${boundAccount.name} (${boundAccount.id}) for API key ${apiKeyData.name}`
           )
           // 更新账户的最后使用时间
-          if (accountType === 'openai') {
-            await openaiAccountService.recordUsage(boundAccount.id, 0)
-          } else {
-            await openaiResponsesAccountService.updateAccount(boundAccount.id, {
-              lastUsedAt: new Date().toISOString()
-            })
-          }
+          await this.updateAccountLastUsed(boundAccount.id, accountType)
           return {
             accountId: boundAccount.id,
             accountType
@@ -292,7 +306,7 @@ class UnifiedOpenAIScheduler {
               `🎯 Using sticky session account: ${mappedAccount.accountId} (${mappedAccount.accountType}) for session ${sessionHash}`
             )
             // 更新账户的最后使用时间
-            await openaiAccountService.recordUsage(mappedAccount.accountId, 0)
+            await this.updateAccountLastUsed(mappedAccount.accountId, mappedAccount.accountType)
             return mappedAccount
           } else {
             logger.warn(
@@ -321,12 +335,8 @@ class UnifiedOpenAIScheduler {
         }
       }
 
-      // 按最后使用时间排序（最久未使用的优先，与 Claude 保持一致）
-      const sortedAccounts = availableAccounts.sort((a, b) => {
-        const aLastUsed = new Date(a.lastUsedAt || 0).getTime()
-        const bLastUsed = new Date(b.lastUsedAt || 0).getTime()
-        return aLastUsed - bLastUsed // 最久未使用的优先
-      })
+      // 按优先级和最后使用时间排序（与 Claude/Gemini 调度保持一致）
+      const sortedAccounts = this._sortAccountsByPriority(availableAccounts)
 
       // 选择第一个账户
       const selectedAccount = sortedAccounts[0]
@@ -344,11 +354,11 @@ class UnifiedOpenAIScheduler {
       }
 
       logger.info(
-        `🎯 Selected account: ${selectedAccount.name} (${selectedAccount.accountId}, ${selectedAccount.accountType}) for API key ${apiKeyData.name}`
+        `🎯 Selected account: ${selectedAccount.name} (${selectedAccount.accountId}, ${selectedAccount.accountType}, priority: ${selectedAccount.priority || 50}) for API key ${apiKeyData.name}`
       )
 
       // 更新账户的最后使用时间
-      await openaiAccountService.recordUsage(selectedAccount.accountId, 0)
+      await this.updateAccountLastUsed(selectedAccount.accountId, selectedAccount.accountType)
 
       return {
         accountId: selectedAccount.accountId,
@@ -494,21 +504,6 @@ class UnifiedOpenAIScheduler {
     return availableAccounts
   }
 
-  // 🔢 按优先级和最后使用时间排序账户（已废弃，改为与 Claude 保持一致，只按最后使用时间排序）
-  // _sortAccountsByPriority(accounts) {
-  //   return accounts.sort((a, b) => {
-  //     // 首先按优先级排序（数字越小优先级越高）
-  //     if (a.priority !== b.priority) {
-  //       return a.priority - b.priority
-  //     }
-
-  //     // 优先级相同时，按最后使用时间排序（最久未使用的优先）
-  //     const aLastUsed = new Date(a.lastUsedAt || 0).getTime()
-  //     const bLastUsed = new Date(b.lastUsedAt || 0).getTime()
-  //     return aLastUsed - bLastUsed
-  //   })
-  // }
-
   // 🔍 检查账户是否可用
   async _isAccountAvailable(accountId, accountType) {
     try {
@@ -817,7 +812,7 @@ class UnifiedOpenAIScheduler {
                 `🎯 Using sticky session account from group: ${mappedAccount.accountId} (${mappedAccount.accountType})`
               )
               // 更新账户的最后使用时间
-              await openaiAccountService.recordUsage(mappedAccount.accountId, 0)
+              await this.updateAccountLastUsed(mappedAccount.accountId, mappedAccount.accountType)
               return mappedAccount
             }
           }
@@ -909,12 +904,8 @@ class UnifiedOpenAIScheduler {
         throw error
       }
 
-      // 按最后使用时间排序（最久未使用的优先，与 Claude 保持一致）
-      const sortedAccounts = availableAccounts.sort((a, b) => {
-        const aLastUsed = new Date(a.lastUsedAt || 0).getTime()
-        const bLastUsed = new Date(b.lastUsedAt || 0).getTime()
-        return aLastUsed - bLastUsed // 最久未使用的优先
-      })
+      // 按优先级和最后使用时间排序（与 Claude/Gemini 调度保持一致）
+      const sortedAccounts = this._sortAccountsByPriority(availableAccounts)
 
       // 选择第一个账户
       const selectedAccount = sortedAccounts[0]
@@ -932,11 +923,11 @@ class UnifiedOpenAIScheduler {
       }
 
       logger.info(
-        `🎯 Selected account from group: ${selectedAccount.name} (${selectedAccount.accountId})`
+        `🎯 Selected account from group: ${selectedAccount.name} (${selectedAccount.accountId}, ${selectedAccount.accountType}, priority: ${selectedAccount.priority || 50})`
       )
 
       // 更新账户的最后使用时间
-      await openaiAccountService.recordUsage(selectedAccount.accountId, 0)
+      await this.updateAccountLastUsed(selectedAccount.accountId, selectedAccount.accountType)
 
       return {
         accountId: selectedAccount.accountId,
@@ -958,9 +949,12 @@ class UnifiedOpenAIScheduler {
   async updateAccountLastUsed(accountId, accountType) {
     try {
       if (accountType === 'openai') {
-        await openaiAccountService.updateAccount(accountId, {
-          lastUsedAt: new Date().toISOString()
-        })
+        await openaiAccountService.recordUsage(accountId, 0)
+        return
+      }
+
+      if (accountType === 'openai-responses') {
+        await openaiResponsesAccountService.recordUsage(accountId, 0)
       }
     } catch (error) {
       logger.warn(`⚠️ Failed to update last used time for account ${accountId}:`, error)

src/services/userMessageQueueService.js

@@ -121,12 +121,23 @@ class UserMessageQueueService {
    * @param {string} accountId - 账户ID
    * @param {string} requestId - 请求ID（可选，会自动生成）
    * @param {number} timeoutMs - 超时时间（可选，使用配置默认值）
+   * @param {Object} accountConfig - 账户级配置（可选），优先级高于全局配置
+   * @param {number} accountConfig.maxConcurrency - 账户级串行队列开关：>0启用，=0使用全局配置
    * @returns {Promise<{acquired: boolean, requestId: string, error?: string}>}
    */
-  async acquireQueueLock(accountId, requestId = null, timeoutMs = null) {
+  async acquireQueueLock(accountId, requestId = null, timeoutMs = null, accountConfig = null) {
     const cfg = await this.getConfig()
 
-    if (!cfg.enabled) {
+    // 账户级配置优先：maxConcurrency > 0 时强制启用，忽略全局开关
+    let queueEnabled = cfg.enabled
+    if (accountConfig && accountConfig.maxConcurrency > 0) {
+      queueEnabled = true
+      logger.debug(
+        `📬 User message queue: account-level queue enabled for account ${accountId} (maxConcurrency=${accountConfig.maxConcurrency})`
+      )
+    }
+
+    if (!queueEnabled) {
       return { acquired: true, requestId: requestId || uuidv4(), skipped: true }
     }
 

src/utils/statsHelper.js

@@ -0,0 +1,105 @@
+/**
+ * 统计计算工具函数
+ * 提供百分位数计算、等待时间统计等通用统计功能
+ */
+
+/**
+ * 计算百分位数（使用 nearest-rank 方法）
+ * @param {number[]} sortedArray - 已排序的数组（升序）
+ * @param {number} percentile - 百分位数 (0-100)
+ * @returns {number} 百分位值
+ *
+ * 边界情况说明：
+ * - percentile=0: 返回最小值 (index=0)
+ * - percentile=100: 返回最大值 (index=len-1)
+ * - percentile=50 且 len=2: 返回第一个元素（nearest-rank 向下取）
+ *
+ * 算法说明（nearest-rank 方法）：
+ * - index = ceil(percentile / 100 * len) - 1
+ * - 示例：len=100, P50 → ceil(50) - 1 = 49（第50个元素，0-indexed）
+ * - 示例：len=100, P99 → ceil(99) - 1 = 98（第99个元素）
+ */
+function getPercentile(sortedArray, percentile) {
+  const len = sortedArray.length
+  if (len === 0) {
+    return 0
+  }
+  if (len === 1) {
+    return sortedArray[0]
+  }
+
+  // 边界处理：percentile <= 0 返回最小值
+  if (percentile <= 0) {
+    return sortedArray[0]
+  }
+  // 边界处理：percentile >= 100 返回最大值
+  if (percentile >= 100) {
+    return sortedArray[len - 1]
+  }
+
+  const index = Math.ceil((percentile / 100) * len) - 1
+  return sortedArray[index]
+}
+
+/**
+ * 计算等待时间分布统计
+ * @param {number[]} waitTimes - 等待时间数组（无需预先排序）
+ * @returns {Object|null} 统计对象，空数组返回 null
+ *
+ * 返回对象包含：
+ * - sampleCount: 样本数量（始终包含，便于调用方判断可靠性）
+ * - count: 样本数量（向后兼容）
+ * - min: 最小值
+ * - max: 最大值
+ * - avg: 平均值（四舍五入）
+ * - p50: 50百分位数（中位数）
+ * - p90: 90百分位数
+ * - p99: 99百分位数
+ * - sampleSizeWarning: 样本量不足时的警告信息（样本 < 10）
+ * - p90Unreliable: P90 统计不可靠标记（样本 < 10）
+ * - p99Unreliable: P99 统计不可靠标记（样本 < 100）
+ *
+ * 可靠性标记说明（详见 design.md Decision 6）：
+ * - 样本 < 10: P90 和 P99 都不可靠
+ * - 样本 < 100: P99 不可靠（P90 需要 10 个样本，P99 需要 100 个样本）
+ * - 即使标记为不可靠，仍返回计算值供参考
+ */
+function calculateWaitTimeStats(waitTimes) {
+  if (!waitTimes || waitTimes.length === 0) {
+    return null
+  }
+
+  const sorted = [...waitTimes].sort((a, b) => a - b)
+  const sum = sorted.reduce((a, b) => a + b, 0)
+  const len = sorted.length
+
+  const stats = {
+    sampleCount: len, // 新增：始终包含样本数
+    count: len, // 向后兼容
+    min: sorted[0],
+    max: sorted[len - 1],
+    avg: Math.round(sum / len),
+    p50: getPercentile(sorted, 50),
+    p90: getPercentile(sorted, 90),
+    p99: getPercentile(sorted, 99)
+  }
+
+  // 渐进式可靠性标记（详见 design.md Decision 6）
+  // 样本 < 10: P90 不可靠（P90 至少需要 ceil(100/10) = 10 个样本）
+  if (len < 10) {
+    stats.sampleSizeWarning = 'Results may be inaccurate due to small sample size'
+    stats.p90Unreliable = true
+  }
+
+  // 样本 < 100: P99 不可靠（P99 至少需要 ceil(100/1) = 100 个样本）
+  if (len < 100) {
+    stats.p99Unreliable = true
+  }
+
+  return stats
+}
+
+module.exports = {
+  getPercentile,
+  calculateWaitTimeStats
+}

src/utils/streamHelper.js

@@ -0,0 +1,36 @@
+/**
+ * Stream Helper Utilities
+ * 流处理辅助工具函数
+ */
+
+/**
+ * 检查响应流是否仍然可写（客户端连接是否有效）
+ * @param {import('http').ServerResponse} stream - HTTP响应流
+ * @returns {boolean} 如果流可写返回true，否则返回false
+ */
+function isStreamWritable(stream) {
+  if (!stream) {
+    return false
+  }
+
+  // 检查流是否已销毁
+  if (stream.destroyed) {
+    return false
+  }
+
+  // 检查底层socket是否已销毁
+  if (stream.socket?.destroyed) {
+    return false
+  }
+
+  // 检查流是否已结束写入
+  if (stream.writableEnded) {
+    return false
+  }
+
+  return true
+}
+
+module.exports = {
+  isStreamWritable
+}

src/utils/warmupInterceptor.js

@@ -0,0 +1,202 @@
+'use strict'
+
+const { v4: uuidv4 } = require('uuid')
+
+/**
+ * 预热请求拦截器
+ * 检测并拦截低价值请求（标题生成、Warmup等），直接返回模拟响应
+ */
+
+/**
+ * 检测是否为预热请求
+ * @param {Object} body - 请求体
+ * @returns {boolean}
+ */
+function isWarmupRequest(body) {
+  if (!body) {
+    return false
+  }
+
+  // 检查 messages
+  if (body.messages && Array.isArray(body.messages)) {
+    for (const msg of body.messages) {
+      // 处理 content 为数组的情况
+      if (Array.isArray(msg.content)) {
+        for (const content of msg.content) {
+          if (content.type === 'text' && typeof content.text === 'string') {
+            if (isTitleOrWarmupText(content.text)) {
+              return true
+            }
+          }
+        }
+      }
+      // 处理 content 为字符串的情况
+      if (typeof msg.content === 'string') {
+        if (isTitleOrWarmupText(msg.content)) {
+          return true
+        }
+      }
+    }
+  }
+
+  // 检查 system prompt
+  if (body.system) {
+    const systemText = extractSystemText(body.system)
+    if (isTitleExtractionSystemPrompt(systemText)) {
+      return true
+    }
+  }
+
+  return false
+}
+
+/**
+ * 检查文本是否为标题生成或Warmup请求
+ */
+function isTitleOrWarmupText(text) {
+  if (!text) {
+    return false
+  }
+  return (
+    text.includes('Please write a 5-10 word title for the following conversation:') ||
+    text === 'Warmup'
+  )
+}
+
+/**
+ * 检查system prompt是否为标题提取类型
+ */
+function isTitleExtractionSystemPrompt(systemText) {
+  if (!systemText) {
+    return false
+  }
+  return systemText.includes(
+    'nalyze if this message indicates a new conversation topic. If it does, extract a 2-3 word title'
+  )
+}
+
+/**
+ * 从system字段提取文本
+ */
+function extractSystemText(system) {
+  if (typeof system === 'string') {
+    return system
+  }
+  if (Array.isArray(system)) {
+    return system.map((s) => (typeof s === 'object' ? s.text || '' : String(s))).join('')
+  }
+  return ''
+}
+
+/**
+ * 生成模拟的非流式响应
+ * @param {string} model - 模型名称
+ * @returns {Object}
+ */
+function buildMockWarmupResponse(model) {
+  return {
+    id: `msg_warmup_${uuidv4().replace(/-/g, '').slice(0, 20)}`,
+    type: 'message',
+    role: 'assistant',
+    content: [{ type: 'text', text: 'New Conversation' }],
+    model: model || 'claude-3-5-sonnet-20241022',
+    stop_reason: 'end_turn',
+    stop_sequence: null,
+    usage: {
+      input_tokens: 10,
+      output_tokens: 2
+    }
+  }
+}
+
+/**
+ * 发送模拟的流式响应
+ * @param {Object} res - Express response对象
+ * @param {string} model - 模型名称
+ */
+function sendMockWarmupStream(res, model) {
+  const effectiveModel = model || 'claude-3-5-sonnet-20241022'
+  const messageId = `msg_warmup_${uuidv4().replace(/-/g, '').slice(0, 20)}`
+
+  const events = [
+    {
+      event: 'message_start',
+      data: {
+        message: {
+          content: [],
+          id: messageId,
+          model: effectiveModel,
+          role: 'assistant',
+          stop_reason: null,
+          stop_sequence: null,
+          type: 'message',
+          usage: { input_tokens: 10, output_tokens: 0 }
+        },
+        type: 'message_start'
+      }
+    },
+    {
+      event: 'content_block_start',
+      data: {
+        content_block: { text: '', type: 'text' },
+        index: 0,
+        type: 'content_block_start'
+      }
+    },
+    {
+      event: 'content_block_delta',
+      data: {
+        delta: { text: 'New', type: 'text_delta' },
+        index: 0,
+        type: 'content_block_delta'
+      }
+    },
+    {
+      event: 'content_block_delta',
+      data: {
+        delta: { text: ' Conversation', type: 'text_delta' },
+        index: 0,
+        type: 'content_block_delta'
+      }
+    },
+    {
+      event: 'content_block_stop',
+      data: { index: 0, type: 'content_block_stop' }
+    },
+    {
+      event: 'message_delta',
+      data: {
+        delta: { stop_reason: 'end_turn', stop_sequence: null },
+        type: 'message_delta',
+        usage: { input_tokens: 10, output_tokens: 2 }
+      }
+    },
+    {
+      event: 'message_stop',
+      data: { type: 'message_stop' }
+    }
+  ]
+
+  let index = 0
+  const sendNext = () => {
+    if (index >= events.length) {
+      res.end()
+      return
+    }
+
+    const { event, data } = events[index]
+    res.write(`event: ${event}\ndata: ${JSON.stringify(data)}\n\n`)
+    index++
+
+    // 模拟网络延迟
+    setTimeout(sendNext, 20)
+  }
+
+  sendNext()
+}
+
+module.exports = {
+  isWarmupRequest,
+  buildMockWarmupResponse,
+  sendMockWarmupStream
+}

tests/concurrencyQueue.integration.test.js

@@ -0,0 +1,860 @@
+/**
+ * 并发请求排队功能集成测试
+ *
+ * 测试分为三个层次：
+ * 1. Mock 测试 - 测试核心逻辑，不需要真实 Redis
+ * 2. Redis 方法测试 - 测试 Redis 操作的原子性和正确性
+ * 3. 端到端场景测试 - 测试完整的排队流程
+ *
+ * 运行方式：
+ * - npm test -- concurrencyQueue.integration  # 运行所有测试（Mock 部分）
+ * - REDIS_TEST=1 npm test -- concurrencyQueue.integration  # 包含真实 Redis 测试
+ */
+
+// Mock logger to avoid console output during tests
+jest.mock('../src/utils/logger', () => ({
+  api: jest.fn(),
+  warn: jest.fn(),
+  error: jest.fn(),
+  info: jest.fn(),
+  database: jest.fn(),
+  security: jest.fn()
+}))
+
+const redis = require('../src/models/redis')
+const claudeRelayConfigService = require('../src/services/claudeRelayConfigService')
+
+// Helper: sleep function
+const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms))
+
+// Helper: 创建模拟的 req/res 对象
+function createMockReqRes() {
+  const listeners = {}
+  const req = {
+    destroyed: false,
+    once: jest.fn((event, handler) => {
+      listeners[`req:${event}`] = handler
+    }),
+    removeListener: jest.fn((event) => {
+      delete listeners[`req:${event}`]
+    }),
+    // 触发事件的辅助方法
+    emit: (event) => {
+      const handler = listeners[`req:${event}`]
+      if (handler) {
+        handler()
+      }
+    }
+  }
+
+  const res = {
+    once: jest.fn((event, handler) => {
+      listeners[`res:${event}`] = handler
+    }),
+    removeListener: jest.fn((event) => {
+      delete listeners[`res:${event}`]
+    }),
+    emit: (event) => {
+      const handler = listeners[`res:${event}`]
+      if (handler) {
+        handler()
+      }
+    }
+  }
+
+  return { req, res, listeners }
+}
+
+// ============================================
+// 第一部分：Mock 测试 - waitForConcurrencySlot 核心逻辑
+// ============================================
+describe('ConcurrencyQueue Integration Tests', () => {
+  describe('Part 1: waitForConcurrencySlot Logic (Mocked)', () => {
+    // 导入 auth 模块中的 waitForConcurrencySlot
+    // 由于它是内部函数，我们需要通过测试其行为来验证
+    // 这里我们模拟整个流程
+
+    let mockRedis
+
+    beforeEach(() => {
+      jest.clearAllMocks()
+
+      // 创建 Redis mock
+      mockRedis = {
+        concurrencyCount: {},
+        queueCount: {},
+        stats: {},
+        waitTimes: {},
+        globalWaitTimes: []
+      }
+
+      // Mock Redis 并发方法
+      jest.spyOn(redis, 'incrConcurrency').mockImplementation(async (keyId, requestId, _lease) => {
+        if (!mockRedis.concurrencyCount[keyId]) {
+          mockRedis.concurrencyCount[keyId] = new Set()
+        }
+        mockRedis.concurrencyCount[keyId].add(requestId)
+        return mockRedis.concurrencyCount[keyId].size
+      })
+
+      jest.spyOn(redis, 'decrConcurrency').mockImplementation(async (keyId, requestId) => {
+        if (mockRedis.concurrencyCount[keyId]) {
+          mockRedis.concurrencyCount[keyId].delete(requestId)
+          return mockRedis.concurrencyCount[keyId].size
+        }
+        return 0
+      })
+
+      // Mock 排队计数方法
+      jest.spyOn(redis, 'incrConcurrencyQueue').mockImplementation(async (keyId) => {
+        mockRedis.queueCount[keyId] = (mockRedis.queueCount[keyId] || 0) + 1
+        return mockRedis.queueCount[keyId]
+      })
+
+      jest.spyOn(redis, 'decrConcurrencyQueue').mockImplementation(async (keyId) => {
+        mockRedis.queueCount[keyId] = Math.max(0, (mockRedis.queueCount[keyId] || 0) - 1)
+        return mockRedis.queueCount[keyId]
+      })
+
+      jest
+        .spyOn(redis, 'getConcurrencyQueueCount')
+        .mockImplementation(async (keyId) => mockRedis.queueCount[keyId] || 0)
+
+      // Mock 统计方法
+      jest.spyOn(redis, 'incrConcurrencyQueueStats').mockImplementation(async (keyId, field) => {
+        if (!mockRedis.stats[keyId]) {
+          mockRedis.stats[keyId] = {}
+        }
+        mockRedis.stats[keyId][field] = (mockRedis.stats[keyId][field] || 0) + 1
+        return mockRedis.stats[keyId][field]
+      })
+
+      jest.spyOn(redis, 'recordQueueWaitTime').mockResolvedValue(undefined)
+      jest.spyOn(redis, 'recordGlobalQueueWaitTime').mockResolvedValue(undefined)
+    })
+
+    afterEach(() => {
+      jest.restoreAllMocks()
+    })
+
+    describe('Slot Acquisition Flow', () => {
+      it('should acquire slot immediately when under concurrency limit', async () => {
+        // 模拟 waitForConcurrencySlot 的行为
+        const keyId = 'test-key-1'
+        const requestId = 'req-1'
+        const concurrencyLimit = 5
+
+        // 直接测试 incrConcurrency 的行为
+        const count = await redis.incrConcurrency(keyId, requestId, 300)
+
+        expect(count).toBe(1)
+        expect(count).toBeLessThanOrEqual(concurrencyLimit)
+      })
+
+      it('should track multiple concurrent requests correctly', async () => {
+        const keyId = 'test-key-2'
+        const concurrencyLimit = 3
+
+        // 模拟多个并发请求
+        const results = []
+        for (let i = 1; i <= 5; i++) {
+          const count = await redis.incrConcurrency(keyId, `req-${i}`, 300)
+          results.push({ requestId: `req-${i}`, count, exceeds: count > concurrencyLimit })
+        }
+
+        // 前3个应该在限制内
+        expect(results[0].exceeds).toBe(false)
+        expect(results[1].exceeds).toBe(false)
+        expect(results[2].exceeds).toBe(false)
+        // 后2个超过限制
+        expect(results[3].exceeds).toBe(true)
+        expect(results[4].exceeds).toBe(true)
+      })
+
+      it('should release slot and allow next request', async () => {
+        const keyId = 'test-key-3'
+        const concurrencyLimit = 1
+
+        // 第一个请求获取槽位
+        const count1 = await redis.incrConcurrency(keyId, 'req-1', 300)
+        expect(count1).toBe(1)
+
+        // 第二个请求超限
+        const count2 = await redis.incrConcurrency(keyId, 'req-2', 300)
+        expect(count2).toBe(2)
+        expect(count2).toBeGreaterThan(concurrencyLimit)
+
+        // 释放第二个请求（因为超限）
+        await redis.decrConcurrency(keyId, 'req-2')
+
+        // 释放第一个请求
+        await redis.decrConcurrency(keyId, 'req-1')
+
+        // 现在第三个请求应该能获取
+        const count3 = await redis.incrConcurrency(keyId, 'req-3', 300)
+        expect(count3).toBe(1)
+      })
+    })
+
+    describe('Queue Count Management', () => {
+      it('should increment and decrement queue count atomically', async () => {
+        const keyId = 'test-key-4'
+
+        // 增加排队计数
+        const count1 = await redis.incrConcurrencyQueue(keyId, 60000)
+        expect(count1).toBe(1)
+
+        const count2 = await redis.incrConcurrencyQueue(keyId, 60000)
+        expect(count2).toBe(2)
+
+        // 减少排队计数
+        const count3 = await redis.decrConcurrencyQueue(keyId)
+        expect(count3).toBe(1)
+
+        const count4 = await redis.decrConcurrencyQueue(keyId)
+        expect(count4).toBe(0)
+      })
+
+      it('should not go below zero on decrement', async () => {
+        const keyId = 'test-key-5'
+
+        // 直接减少（没有先增加）
+        const count = await redis.decrConcurrencyQueue(keyId)
+        expect(count).toBe(0)
+      })
+
+      it('should handle concurrent queue operations', async () => {
+        const keyId = 'test-key-6'
+
+        // 并发增加
+        const increments = await Promise.all([
+          redis.incrConcurrencyQueue(keyId, 60000),
+          redis.incrConcurrencyQueue(keyId, 60000),
+          redis.incrConcurrencyQueue(keyId, 60000)
+        ])
+
+        // 所有增量应该是连续的
+        const sortedIncrements = [...increments].sort((a, b) => a - b)
+        expect(sortedIncrements).toEqual([1, 2, 3])
+      })
+    })
+
+    describe('Statistics Tracking', () => {
+      it('should track entered/success/timeout/cancelled stats', async () => {
+        const keyId = 'test-key-7'
+
+        await redis.incrConcurrencyQueueStats(keyId, 'entered')
+        await redis.incrConcurrencyQueueStats(keyId, 'entered')
+        await redis.incrConcurrencyQueueStats(keyId, 'success')
+        await redis.incrConcurrencyQueueStats(keyId, 'timeout')
+        await redis.incrConcurrencyQueueStats(keyId, 'cancelled')
+
+        expect(mockRedis.stats[keyId]).toEqual({
+          entered: 2,
+          success: 1,
+          timeout: 1,
+          cancelled: 1
+        })
+      })
+    })
+
+    describe('Client Disconnection Handling', () => {
+      it('should detect client disconnection via close event', async () => {
+        const { req } = createMockReqRes()
+
+        let clientDisconnected = false
+
+        // 设置监听器
+        req.once('close', () => {
+          clientDisconnected = true
+        })
+
+        // 模拟客户端断开
+        req.emit('close')
+
+        expect(clientDisconnected).toBe(true)
+      })
+
+      it('should detect pre-destroyed request', () => {
+        const { req } = createMockReqRes()
+        req.destroyed = true
+
+        expect(req.destroyed).toBe(true)
+      })
+    })
+
+    describe('Exponential Backoff Simulation', () => {
+      it('should increase poll interval with backoff', () => {
+        const config = {
+          pollIntervalMs: 200,
+          maxPollIntervalMs: 2000,
+          backoffFactor: 1.5,
+          jitterRatio: 0 // 禁用抖动以便测试
+        }
+
+        let interval = config.pollIntervalMs
+        const intervals = [interval]
+
+        for (let i = 0; i < 5; i++) {
+          interval = Math.min(interval * config.backoffFactor, config.maxPollIntervalMs)
+          intervals.push(interval)
+        }
+
+        // 验证指数增长
+        expect(intervals[1]).toBe(300) // 200 * 1.5
+        expect(intervals[2]).toBe(450) // 300 * 1.5
+        expect(intervals[3]).toBe(675) // 450 * 1.5
+        expect(intervals[4]).toBe(1012.5) // 675 * 1.5
+        expect(intervals[5]).toBe(1518.75) // 1012.5 * 1.5
+      })
+
+      it('should cap interval at maximum', () => {
+        const config = {
+          pollIntervalMs: 1000,
+          maxPollIntervalMs: 2000,
+          backoffFactor: 1.5
+        }
+
+        let interval = config.pollIntervalMs
+
+        for (let i = 0; i < 10; i++) {
+          interval = Math.min(interval * config.backoffFactor, config.maxPollIntervalMs)
+        }
+
+        expect(interval).toBe(2000)
+      })
+
+      it('should apply jitter within expected range', () => {
+        const baseInterval = 1000
+        const jitterRatio = 0.2 // ±20%
+        const results = []
+
+        for (let i = 0; i < 100; i++) {
+          const randomValue = Math.random()
+          const jitter = baseInterval * jitterRatio * (randomValue * 2 - 1)
+          const finalInterval = baseInterval + jitter
+          results.push(finalInterval)
+        }
+
+        const min = Math.min(...results)
+        const max = Math.max(...results)
+
+        // 所有结果应该在 [800, 1200] 范围内
+        expect(min).toBeGreaterThanOrEqual(800)
+        expect(max).toBeLessThanOrEqual(1200)
+      })
+    })
+  })
+
+  // ============================================
+  // 第二部分：并发竞争场景测试
+  // ============================================
+  describe('Part 2: Concurrent Race Condition Tests', () => {
+    beforeEach(() => {
+      jest.clearAllMocks()
+    })
+
+    afterEach(() => {
+      jest.restoreAllMocks()
+    })
+
+    describe('Race Condition: Multiple Requests Competing for Same Slot', () => {
+      it('should handle race condition when multiple requests try to acquire last slot', async () => {
+        const keyId = 'race-test-1'
+        const concurrencyLimit = 1
+        const concurrencyState = { count: 0, holders: new Set() }
+
+        // 模拟原子的 incrConcurrency
+        jest.spyOn(redis, 'incrConcurrency').mockImplementation(async (key, reqId) => {
+          // 模拟原子操作
+          concurrencyState.count++
+          concurrencyState.holders.add(reqId)
+          return concurrencyState.count
+        })
+
+        jest.spyOn(redis, 'decrConcurrency').mockImplementation(async (key, reqId) => {
+          if (concurrencyState.holders.has(reqId)) {
+            concurrencyState.count--
+            concurrencyState.holders.delete(reqId)
+          }
+          return concurrencyState.count
+        })
+
+        // 5个请求同时竞争1个槽位
+        const requests = Array.from({ length: 5 }, (_, i) => `req-${i + 1}`)
+
+        const acquireResults = await Promise.all(
+          requests.map(async (reqId) => {
+            const count = await redis.incrConcurrency(keyId, reqId, 300)
+            const acquired = count <= concurrencyLimit
+
+            if (!acquired) {
+              // 超限，释放
+              await redis.decrConcurrency(keyId, reqId)
+            }
+
+            return { reqId, count, acquired }
+          })
+        )
+
+        // 只有一个请求应该成功获取槽位
+        const successfulAcquires = acquireResults.filter((r) => r.acquired)
+        expect(successfulAcquires.length).toBe(1)
+
+        // 最终并发计数应该是1
+        expect(concurrencyState.count).toBe(1)
+      })
+
+      it('should maintain consistency under high contention', async () => {
+        const keyId = 'race-test-2'
+        const concurrencyLimit = 3
+        const requestCount = 20
+        const concurrencyState = { count: 0, maxSeen: 0 }
+
+        jest.spyOn(redis, 'incrConcurrency').mockImplementation(async () => {
+          concurrencyState.count++
+          concurrencyState.maxSeen = Math.max(concurrencyState.maxSeen, concurrencyState.count)
+          return concurrencyState.count
+        })
+
+        jest.spyOn(redis, 'decrConcurrency').mockImplementation(async () => {
+          concurrencyState.count = Math.max(0, concurrencyState.count - 1)
+          return concurrencyState.count
+        })
+
+        // 模拟多轮请求
+        const activeRequests = []
+
+        for (let i = 0; i < requestCount; i++) {
+          const count = await redis.incrConcurrency(keyId, `req-${i}`, 300)
+
+          if (count <= concurrencyLimit) {
+            activeRequests.push(`req-${i}`)
+
+            // 模拟处理时间后释放
+            setTimeout(async () => {
+              await redis.decrConcurrency(keyId, `req-${i}`)
+            }, Math.random() * 50)
+          } else {
+            await redis.decrConcurrency(keyId, `req-${i}`)
+          }
+
+          // 随机延迟
+          await sleep(Math.random() * 10)
+        }
+
+        // 等待所有请求完成
+        await sleep(100)
+
+        // 最大并发不应超过限制
+        expect(concurrencyState.maxSeen).toBeLessThanOrEqual(concurrencyLimit + requestCount) // 允许短暂超限
+      })
+    })
+
+    describe('Queue Overflow Protection', () => {
+      it('should reject requests when queue is full', async () => {
+        const keyId = 'overflow-test-1'
+        const maxQueueSize = 5
+        const queueState = { count: 0 }
+
+        jest.spyOn(redis, 'incrConcurrencyQueue').mockImplementation(async () => {
+          queueState.count++
+          return queueState.count
+        })
+
+        jest.spyOn(redis, 'decrConcurrencyQueue').mockImplementation(async () => {
+          queueState.count = Math.max(0, queueState.count - 1)
+          return queueState.count
+        })
+
+        const results = []
+
+        // 尝试10个请求进入队列
+        for (let i = 0; i < 10; i++) {
+          const queueCount = await redis.incrConcurrencyQueue(keyId, 60000)
+
+          if (queueCount > maxQueueSize) {
+            // 队列满，释放并拒绝
+            await redis.decrConcurrencyQueue(keyId)
+            results.push({ index: i, accepted: false })
+          } else {
+            results.push({ index: i, accepted: true, position: queueCount })
+          }
+        }
+
+        const accepted = results.filter((r) => r.accepted)
+        const rejected = results.filter((r) => !r.accepted)
+
+        expect(accepted.length).toBe(5)
+        expect(rejected.length).toBe(5)
+      })
+    })
+  })
+
+  // ============================================
+  // 第三部分：真实 Redis 集成测试（可选）
+  // ============================================
+  describe('Part 3: Real Redis Integration Tests', () => {
+    const skipRealRedis = !process.env.REDIS_TEST
+
+    // 辅助函数：检查 Redis 连接
+    async function checkRedisConnection() {
+      try {
+        const client = redis.getClient()
+        if (!client) {
+          return false
+        }
+        await client.ping()
+        return true
+      } catch {
+        return false
+      }
+    }
+
+    beforeAll(async () => {
+      if (skipRealRedis) {
+        console.log('⏭️  Skipping real Redis tests (set REDIS_TEST=1 to enable)')
+        return
+      }
+
+      const connected = await checkRedisConnection()
+      if (!connected) {
+        console.log('⚠️  Redis not connected, skipping real Redis tests')
+      }
+    })
+
+    // 清理测试数据
+    afterEach(async () => {
+      if (skipRealRedis) {
+        return
+      }
+
+      try {
+        const client = redis.getClient()
+        if (!client) {
+          return
+        }
+
+        // 清理测试键
+        const testKeys = await client.keys('concurrency:queue:test-*')
+        if (testKeys.length > 0) {
+          await client.del(...testKeys)
+        }
+      } catch {
+        // 忽略清理错误
+      }
+    })
+
+    describe('Redis Queue Operations', () => {
+      const testOrSkip = skipRealRedis ? it.skip : it
+
+      testOrSkip('should atomically increment queue count with TTL', async () => {
+        const keyId = 'test-redis-queue-1'
+        const timeoutMs = 5000
+
+        const count1 = await redis.incrConcurrencyQueue(keyId, timeoutMs)
+        expect(count1).toBe(1)
+
+        const count2 = await redis.incrConcurrencyQueue(keyId, timeoutMs)
+        expect(count2).toBe(2)
+
+        // 验证 TTL 被设置
+        const client = redis.getClient()
+        const ttl = await client.ttl(`concurrency:queue:${keyId}`)
+        expect(ttl).toBeGreaterThan(0)
+        expect(ttl).toBeLessThanOrEqual(Math.ceil(timeoutMs / 1000) + 30)
+      })
+
+      testOrSkip('should atomically decrement and delete when zero', async () => {
+        const keyId = 'test-redis-queue-2'
+
+        await redis.incrConcurrencyQueue(keyId, 60000)
+        const count = await redis.decrConcurrencyQueue(keyId)
+
+        expect(count).toBe(0)
+
+        // 验证键已删除
+        const client = redis.getClient()
+        const exists = await client.exists(`concurrency:queue:${keyId}`)
+        expect(exists).toBe(0)
+      })
+
+      testOrSkip('should handle concurrent increments correctly', async () => {
+        const keyId = 'test-redis-queue-3'
+        const numRequests = 10
+
+        // 并发增加
+        const results = await Promise.all(
+          Array.from({ length: numRequests }, () => redis.incrConcurrencyQueue(keyId, 60000))
+        )
+
+        // 所有结果应该是 1 到 numRequests
+        const sorted = [...results].sort((a, b) => a - b)
+        expect(sorted).toEqual(Array.from({ length: numRequests }, (_, i) => i + 1))
+      })
+    })
+
+    describe('Redis Stats Operations', () => {
+      const testOrSkip = skipRealRedis ? it.skip : it
+
+      testOrSkip('should track queue statistics correctly', async () => {
+        const keyId = 'test-redis-stats-1'
+
+        await redis.incrConcurrencyQueueStats(keyId, 'entered')
+        await redis.incrConcurrencyQueueStats(keyId, 'entered')
+        await redis.incrConcurrencyQueueStats(keyId, 'success')
+        await redis.incrConcurrencyQueueStats(keyId, 'timeout')
+
+        const stats = await redis.getConcurrencyQueueStats(keyId)
+
+        expect(stats.entered).toBe(2)
+        expect(stats.success).toBe(1)
+        expect(stats.timeout).toBe(1)
+        expect(stats.cancelled).toBe(0)
+      })
+
+      testOrSkip('should record and retrieve wait times', async () => {
+        const keyId = 'test-redis-wait-1'
+        const waitTimes = [100, 200, 150, 300, 250]
+
+        for (const wt of waitTimes) {
+          await redis.recordQueueWaitTime(keyId, wt)
+        }
+
+        const recorded = await redis.getQueueWaitTimes(keyId)
+
+        // 应该按 LIFO 顺序存储
+        expect(recorded.length).toBe(5)
+        expect(recorded[0]).toBe(250) // 最后插入的在前面
+      })
+
+      testOrSkip('should record global wait times', async () => {
+        const waitTimes = [500, 600, 700]
+
+        for (const wt of waitTimes) {
+          await redis.recordGlobalQueueWaitTime(wt)
+        }
+
+        const recorded = await redis.getGlobalQueueWaitTimes()
+
+        expect(recorded.length).toBeGreaterThanOrEqual(3)
+      })
+    })
+
+    describe('Redis Cleanup Operations', () => {
+      const testOrSkip = skipRealRedis ? it.skip : it
+
+      testOrSkip('should clear specific queue', async () => {
+        const keyId = 'test-redis-clear-1'
+
+        await redis.incrConcurrencyQueue(keyId, 60000)
+        await redis.incrConcurrencyQueue(keyId, 60000)
+
+        const cleared = await redis.clearConcurrencyQueue(keyId)
+        expect(cleared).toBe(true)
+
+        const count = await redis.getConcurrencyQueueCount(keyId)
+        expect(count).toBe(0)
+      })
+
+      testOrSkip('should clear all queues but preserve stats', async () => {
+        const keyId1 = 'test-redis-clearall-1'
+        const keyId2 = 'test-redis-clearall-2'
+
+        // 创建队列和统计
+        await redis.incrConcurrencyQueue(keyId1, 60000)
+        await redis.incrConcurrencyQueue(keyId2, 60000)
+        await redis.incrConcurrencyQueueStats(keyId1, 'entered')
+
+        // 清理所有队列
+        const cleared = await redis.clearAllConcurrencyQueues()
+        expect(cleared).toBeGreaterThanOrEqual(2)
+
+        // 验证队列已清理
+        const count1 = await redis.getConcurrencyQueueCount(keyId1)
+        const count2 = await redis.getConcurrencyQueueCount(keyId2)
+        expect(count1).toBe(0)
+        expect(count2).toBe(0)
+
+        // 统计应该保留
+        const stats = await redis.getConcurrencyQueueStats(keyId1)
+        expect(stats.entered).toBe(1)
+      })
+    })
+  })
+
+  // ============================================
+  // 第四部分：配置服务集成测试
+  // ============================================
+  describe('Part 4: Configuration Service Integration', () => {
+    beforeEach(() => {
+      // 清除配置缓存
+      claudeRelayConfigService.clearCache()
+    })
+
+    afterEach(() => {
+      jest.restoreAllMocks()
+    })
+
+    describe('Queue Configuration', () => {
+      it('should return default queue configuration', async () => {
+        jest.spyOn(redis, 'getClient').mockReturnValue(null)
+
+        const config = await claudeRelayConfigService.getConfig()
+
+        expect(config.concurrentRequestQueueEnabled).toBe(false)
+        expect(config.concurrentRequestQueueMaxSize).toBe(3)
+        expect(config.concurrentRequestQueueMaxSizeMultiplier).toBe(0)
+        expect(config.concurrentRequestQueueTimeoutMs).toBe(10000)
+      })
+
+      it('should calculate max queue size correctly', async () => {
+        const testCases = [
+          { concurrencyLimit: 5, multiplier: 2, fixedMin: 3, expected: 10 }, // 5*2=10 > 3
+          { concurrencyLimit: 1, multiplier: 1, fixedMin: 5, expected: 5 }, // 1*1=1 < 5
+          { concurrencyLimit: 10, multiplier: 0.5, fixedMin: 3, expected: 5 }, // 10*0.5=5 > 3
+          { concurrencyLimit: 2, multiplier: 1, fixedMin: 10, expected: 10 } // 2*1=2 < 10
+        ]
+
+        for (const tc of testCases) {
+          const maxQueueSize = Math.max(tc.concurrencyLimit * tc.multiplier, tc.fixedMin)
+          expect(maxQueueSize).toBe(tc.expected)
+        }
+      })
+    })
+  })
+
+  // ============================================
+  // 第五部分：端到端场景测试
+  // ============================================
+  describe('Part 5: End-to-End Scenario Tests', () => {
+    describe('Scenario: Claude Code Agent Parallel Tool Calls', () => {
+      it('should handle burst of parallel tool results', async () => {
+        // 模拟 Claude Code Agent 发送多个并行工具结果的场景
+        const concurrencyLimit = 2
+        const maxQueueSize = 5
+
+        const state = {
+          concurrency: 0,
+          queue: 0,
+          completed: 0,
+          rejected: 0
+        }
+
+        // 模拟 8 个并行工具结果请求
+        const requests = Array.from({ length: 8 }, (_, i) => ({
+          id: `tool-result-${i + 1}`,
+          startTime: Date.now()
+        }))
+
+        // 模拟处理逻辑
+        async function processRequest(req) {
+          // 尝试获取并发槽位
+          state.concurrency++
+
+          if (state.concurrency > concurrencyLimit) {
+            // 超限，进入队列
+            state.concurrency--
+            state.queue++
+
+            if (state.queue > maxQueueSize) {
+              // 队列满，拒绝
+              state.queue--
+              state.rejected++
+              return { ...req, status: 'rejected', reason: 'queue_full' }
+            }
+
+            // 等待槽位（模拟）
+            await sleep(Math.random() * 100)
+            state.queue--
+            state.concurrency++
+          }
+
+          // 处理请求
+          await sleep(50) // 模拟处理时间
+          state.concurrency--
+          state.completed++
+
+          return { ...req, status: 'completed', duration: Date.now() - req.startTime }
+        }
+
+        const results = await Promise.all(requests.map(processRequest))
+
+        const completed = results.filter((r) => r.status === 'completed')
+        const rejected = results.filter((r) => r.status === 'rejected')
+
+        // 大部分请求应该完成
+        expect(completed.length).toBeGreaterThan(0)
+        // 可能有一些被拒绝
+        expect(state.rejected).toBe(rejected.length)
+
+        console.log(
+          `  ✓ Completed: ${completed.length}, Rejected: ${rejected.length}, Max concurrent: ${concurrencyLimit}`
+        )
+      })
+    })
+
+    describe('Scenario: Graceful Degradation', () => {
+      it('should fallback when Redis fails', async () => {
+        jest
+          .spyOn(redis, 'incrConcurrencyQueue')
+          .mockRejectedValue(new Error('Redis connection lost'))
+
+        // 模拟降级行为：Redis 失败时直接拒绝而不是崩溃
+        let result
+        try {
+          await redis.incrConcurrencyQueue('fallback-test', 60000)
+          result = { success: true }
+        } catch (error) {
+          // 优雅降级：返回 429 而不是 500
+          result = { success: false, fallback: true, error: error.message }
+        }
+
+        expect(result.fallback).toBe(true)
+        expect(result.error).toContain('Redis')
+      })
+    })
+
+    describe('Scenario: Timeout Behavior', () => {
+      it('should respect queue timeout', async () => {
+        const timeoutMs = 100
+        const startTime = Date.now()
+
+        // 模拟等待超时
+        await new Promise((resolve) => setTimeout(resolve, timeoutMs))
+
+        const elapsed = Date.now() - startTime
+        expect(elapsed).toBeGreaterThanOrEqual(timeoutMs - 10) // 允许 10ms 误差
+      })
+
+      it('should track timeout statistics', async () => {
+        const stats = { entered: 0, success: 0, timeout: 0, cancelled: 0 }
+
+        // 模拟多个请求，部分超时
+        const requests = [
+          { id: 'req-1', willTimeout: false },
+          { id: 'req-2', willTimeout: true },
+          { id: 'req-3', willTimeout: false },
+          { id: 'req-4', willTimeout: true }
+        ]
+
+        for (const req of requests) {
+          stats.entered++
+          if (req.willTimeout) {
+            stats.timeout++
+          } else {
+            stats.success++
+          }
+        }
+
+        expect(stats.entered).toBe(4)
+        expect(stats.success).toBe(2)
+        expect(stats.timeout).toBe(2)
+
+        // 成功率应该是 50%
+        const successRate = (stats.success / stats.entered) * 100
+        expect(successRate).toBe(50)
+      })
+    })
+  })
+})

tests/concurrencyQueue.test.js

@@ -0,0 +1,278 @@
+/**
+ * 并发请求排队功能测试
+ * 测试排队逻辑中的核心算法：百分位数计算、等待时间统计、指数退避等
+ *
+ * 注意：Redis 方法的测试需要集成测试环境，这里主要测试纯算法逻辑
+ */
+
+// Mock logger to avoid console output during tests
+jest.mock('../src/utils/logger', () => ({
+  api: jest.fn(),
+  warn: jest.fn(),
+  error: jest.fn(),
+  info: jest.fn(),
+  database: jest.fn(),
+  security: jest.fn()
+}))
+
+// 使用共享的统计工具函数（与生产代码一致）
+const { getPercentile, calculateWaitTimeStats } = require('../src/utils/statsHelper')
+
+describe('ConcurrencyQueue', () => {
+  describe('Percentile Calculation (nearest-rank method)', () => {
+    // 直接测试共享工具函数，确保与生产代码行为一致
+    it('should return 0 for empty array', () => {
+      expect(getPercentile([], 50)).toBe(0)
+    })
+
+    it('should return single element for single-element array', () => {
+      expect(getPercentile([100], 50)).toBe(100)
+      expect(getPercentile([100], 99)).toBe(100)
+    })
+
+    it('should return min for percentile 0', () => {
+      expect(getPercentile([10, 20, 30, 40, 50], 0)).toBe(10)
+    })
+
+    it('should return max for percentile 100', () => {
+      expect(getPercentile([10, 20, 30, 40, 50], 100)).toBe(50)
+    })
+
+    it('should calculate P50 correctly for len=10', () => {
+      // For [10, 20, 30, 40, 50, 60, 70, 80, 90, 100] (len=10)
+      // P50: ceil(50/100 * 10) - 1 = ceil(5) - 1 = 4 → value at index 4 = 50
+      const arr = [10, 20, 30, 40, 50, 60, 70, 80, 90, 100]
+      expect(getPercentile(arr, 50)).toBe(50)
+    })
+
+    it('should calculate P90 correctly for len=10', () => {
+      // For len=10, P90: ceil(90/100 * 10) - 1 = ceil(9) - 1 = 8 → value at index 8 = 90
+      const arr = [10, 20, 30, 40, 50, 60, 70, 80, 90, 100]
+      expect(getPercentile(arr, 90)).toBe(90)
+    })
+
+    it('should calculate P99 correctly for len=100', () => {
+      // For len=100, P99: ceil(99/100 * 100) - 1 = ceil(99) - 1 = 98
+      const arr = Array.from({ length: 100 }, (_, i) => i + 1)
+      expect(getPercentile(arr, 99)).toBe(99)
+    })
+
+    it('should handle two-element array correctly', () => {
+      // For [10, 20] (len=2)
+      // P50: ceil(50/100 * 2) - 1 = ceil(1) - 1 = 0 → value = 10
+      expect(getPercentile([10, 20], 50)).toBe(10)
+    })
+
+    it('should handle negative percentile as 0', () => {
+      expect(getPercentile([10, 20, 30], -10)).toBe(10)
+    })
+
+    it('should handle percentile > 100 as 100', () => {
+      expect(getPercentile([10, 20, 30], 150)).toBe(30)
+    })
+  })
+
+  describe('Wait Time Stats Calculation', () => {
+    // 直接测试共享工具函数
+    it('should return null for empty array', () => {
+      expect(calculateWaitTimeStats([])).toBeNull()
+    })
+
+    it('should return null for null input', () => {
+      expect(calculateWaitTimeStats(null)).toBeNull()
+    })
+
+    it('should return null for undefined input', () => {
+      expect(calculateWaitTimeStats(undefined)).toBeNull()
+    })
+
+    it('should calculate stats correctly for typical data', () => {
+      const waitTimes = [100, 200, 150, 300, 250, 180, 220, 280, 190, 210]
+      const stats = calculateWaitTimeStats(waitTimes)
+
+      expect(stats.count).toBe(10)
+      expect(stats.min).toBe(100)
+      expect(stats.max).toBe(300)
+      // Sum: 100+150+180+190+200+210+220+250+280+300 = 2080
+      expect(stats.avg).toBe(208)
+      expect(stats.sampleSizeWarning).toBeUndefined()
+    })
+
+    it('should add warning for small sample size (< 10)', () => {
+      const waitTimes = [100, 200, 300]
+      const stats = calculateWaitTimeStats(waitTimes)
+
+      expect(stats.count).toBe(3)
+      expect(stats.sampleSizeWarning).toBe('Results may be inaccurate due to small sample size')
+    })
+
+    it('should handle single value', () => {
+      const stats = calculateWaitTimeStats([500])
+
+      expect(stats.count).toBe(1)
+      expect(stats.min).toBe(500)
+      expect(stats.max).toBe(500)
+      expect(stats.avg).toBe(500)
+      expect(stats.p50).toBe(500)
+      expect(stats.p90).toBe(500)
+      expect(stats.p99).toBe(500)
+    })
+
+    it('should sort input array before calculating', () => {
+      const waitTimes = [500, 100, 300, 200, 400]
+      const stats = calculateWaitTimeStats(waitTimes)
+
+      expect(stats.min).toBe(100)
+      expect(stats.max).toBe(500)
+    })
+
+    it('should not modify original array', () => {
+      const waitTimes = [500, 100, 300]
+      calculateWaitTimeStats(waitTimes)
+
+      expect(waitTimes).toEqual([500, 100, 300])
+    })
+  })
+
+  describe('Exponential Backoff with Jitter', () => {
+    /**
+     * 指数退避计算函数（与 auth.js 中的实现一致）
+     * @param {number} currentInterval - 当前轮询间隔
+     * @param {number} backoffFactor - 退避系数
+     * @param {number} jitterRatio - 抖动比例
+     * @param {number} maxInterval - 最大间隔
+     * @param {number} randomValue - 随机值 [0, 1)，用于确定性测试
+     */
+    function calculateNextInterval(
+      currentInterval,
+      backoffFactor,
+      jitterRatio,
+      maxInterval,
+      randomValue
+    ) {
+      let nextInterval = currentInterval * backoffFactor
+      // 抖动范围：[-jitterRatio, +jitterRatio]
+      const jitter = nextInterval * jitterRatio * (randomValue * 2 - 1)
+      nextInterval = nextInterval + jitter
+      return Math.max(1, Math.min(nextInterval, maxInterval))
+    }
+
+    it('should apply exponential backoff without jitter (randomValue=0.5)', () => {
+      // randomValue = 0.5 gives jitter = 0
+      const next = calculateNextInterval(100, 1.5, 0.2, 1000, 0.5)
+      expect(next).toBe(150) // 100 * 1.5 = 150
+    })
+
+    it('should apply maximum positive jitter (randomValue=1.0)', () => {
+      // randomValue = 1.0 gives maximum positive jitter (+20%)
+      const next = calculateNextInterval(100, 1.5, 0.2, 1000, 1.0)
+      // 100 * 1.5 = 150, jitter = 150 * 0.2 * 1 = 30
+      expect(next).toBe(180) // 150 + 30
+    })
+
+    it('should apply maximum negative jitter (randomValue=0.0)', () => {
+      // randomValue = 0.0 gives maximum negative jitter (-20%)
+      const next = calculateNextInterval(100, 1.5, 0.2, 1000, 0.0)
+      // 100 * 1.5 = 150, jitter = 150 * 0.2 * -1 = -30
+      expect(next).toBe(120) // 150 - 30
+    })
+
+    it('should respect maximum interval', () => {
+      const next = calculateNextInterval(800, 1.5, 0.2, 1000, 1.0)
+      // 800 * 1.5 = 1200, with +20% jitter = 1440, capped at 1000
+      expect(next).toBe(1000)
+    })
+
+    it('should never go below 1ms even with extreme negative jitter', () => {
+      const next = calculateNextInterval(1, 1.0, 0.9, 1000, 0.0)
+      // 1 * 1.0 = 1, jitter = 1 * 0.9 * -1 = -0.9
+      // 1 - 0.9 = 0.1, but Math.max(1, ...) ensures minimum is 1
+      expect(next).toBe(1)
+    })
+
+    it('should handle zero jitter ratio', () => {
+      const next = calculateNextInterval(100, 2.0, 0, 1000, 0.0)
+      expect(next).toBe(200) // Pure exponential, no jitter
+    })
+
+    it('should handle large backoff factor', () => {
+      const next = calculateNextInterval(100, 3.0, 0.1, 1000, 0.5)
+      expect(next).toBe(300) // 100 * 3.0 = 300
+    })
+
+    describe('jitter distribution', () => {
+      it('should produce values in expected range', () => {
+        const results = []
+        // Test with various random values
+        for (let r = 0; r <= 1; r += 0.1) {
+          results.push(calculateNextInterval(100, 1.5, 0.2, 1000, r))
+        }
+        // All values should be between 120 (150 - 30) and 180 (150 + 30)
+        expect(Math.min(...results)).toBeGreaterThanOrEqual(120)
+        expect(Math.max(...results)).toBeLessThanOrEqual(180)
+      })
+    })
+  })
+
+  describe('Queue Size Calculation', () => {
+    /**
+     * 最大排队数计算（与 auth.js 中的实现一致）
+     */
+    function calculateMaxQueueSize(concurrencyLimit, multiplier, fixedMin) {
+      return Math.max(concurrencyLimit * multiplier, fixedMin)
+    }
+
+    it('should use multiplier when result is larger', () => {
+      // concurrencyLimit=10, multiplier=2, fixedMin=5
+      // max(10*2, 5) = max(20, 5) = 20
+      expect(calculateMaxQueueSize(10, 2, 5)).toBe(20)
+    })
+
+    it('should use fixed minimum when multiplier result is smaller', () => {
+      // concurrencyLimit=2, multiplier=1, fixedMin=5
+      // max(2*1, 5) = max(2, 5) = 5
+      expect(calculateMaxQueueSize(2, 1, 5)).toBe(5)
+    })
+
+    it('should handle zero multiplier', () => {
+      // concurrencyLimit=10, multiplier=0, fixedMin=3
+      // max(10*0, 3) = max(0, 3) = 3
+      expect(calculateMaxQueueSize(10, 0, 3)).toBe(3)
+    })
+
+    it('should handle fractional multiplier', () => {
+      // concurrencyLimit=10, multiplier=1.5, fixedMin=5
+      // max(10*1.5, 5) = max(15, 5) = 15
+      expect(calculateMaxQueueSize(10, 1.5, 5)).toBe(15)
+    })
+  })
+
+  describe('TTL Calculation', () => {
+    /**
+     * 排队计数器 TTL 计算（与 redis.js 中的实现一致）
+     */
+    function calculateQueueTtl(timeoutMs, bufferSeconds = 30) {
+      return Math.ceil(timeoutMs / 1000) + bufferSeconds
+    }
+
+    it('should calculate TTL with default buffer', () => {
+      // 60000ms = 60s + 30s buffer = 90s
+      expect(calculateQueueTtl(60000)).toBe(90)
+    })
+
+    it('should round up milliseconds to seconds', () => {
+      // 61500ms = ceil(61.5) = 62s + 30s = 92s
+      expect(calculateQueueTtl(61500)).toBe(92)
+    })
+
+    it('should handle custom buffer', () => {
+      // 30000ms = 30s + 60s buffer = 90s
+      expect(calculateQueueTtl(30000, 60)).toBe(90)
+    })
+
+    it('should handle very short timeout', () => {
+      // 1000ms = 1s + 30s = 31s
+      expect(calculateQueueTtl(1000)).toBe(31)
+    })
+  })
+})

web/admin-spa/src/components/accounts/AccountForm.vue

@@ -1320,10 +1320,10 @@
                       class="rounded-lg bg-blue-100 px-3 py-1 text-xs text-blue-700 transition-colors hover:bg-blue-200 dark:bg-blue-900/30 dark:text-blue-400 dark:hover:bg-blue-900/50"
                       type="button"
                       @click="
-                        addPresetMapping('claude-sonnet-4-20250514', 'claude-sonnet-4-20250514')
+                        addPresetMapping('claude-opus-4-5-20251101', 'claude-opus-4-5-20251101')
                       "
                     >
-                      + Sonnet 4
+                      + Opus 4.5
                     </button>
                     <button
                       class="rounded-lg bg-indigo-100 px-3 py-1 text-xs text-indigo-700 transition-colors hover:bg-indigo-200 dark:bg-indigo-900/30 dark:text-indigo-400 dark:hover:bg-indigo-900/50"
@@ -1334,24 +1334,6 @@
                     >
                       + Sonnet 4.5
                     </button>
-                    <button
-                      class="rounded-lg bg-purple-100 px-3 py-1 text-xs text-purple-700 transition-colors hover:bg-purple-200 dark:bg-purple-900/30 dark:text-purple-400 dark:hover:bg-purple-900/50"
-                      type="button"
-                      @click="
-                        addPresetMapping('claude-opus-4-1-20250805', 'claude-opus-4-1-20250805')
-                      "
-                    >
-                      + Opus 4.1
-                    </button>
-                    <button
-                      class="rounded-lg bg-green-100 px-3 py-1 text-xs text-green-700 transition-colors hover:bg-green-200 dark:bg-green-900/30 dark:text-green-400 dark:hover:bg-green-900/50"
-                      type="button"
-                      @click="
-                        addPresetMapping('claude-3-5-haiku-20241022', 'claude-3-5-haiku-20241022')
-                      "
-                    >
-                      + Haiku 3.5
-                    </button>
                     <button
                       class="rounded-lg bg-emerald-100 px-3 py-1 text-xs text-emerald-700 transition-colors hover:bg-emerald-200 dark:bg-emerald-900/30 dark:text-emerald-400 dark:hover:bg-emerald-900/50"
                       type="button"
@@ -1650,6 +1632,47 @@
               </label>
             </div>
 
+            <!-- Claude 账户级串行队列开关 -->
+            <div v-if="form.platform === 'claude'" class="mt-4">
+              <label class="flex items-start">
+                <input
+                  v-model="form.serialQueueEnabled"
+                  class="mt-1 text-blue-600 focus:ring-blue-500 dark:border-gray-600 dark:bg-gray-700"
+                  type="checkbox"
+                />
+                <div class="ml-3">
+                  <span class="text-sm font-medium text-gray-700 dark:text-gray-300">
+                    启用账户级串行队列
+                  </span>
+                  <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+                    开启后强制该账户的用户消息串行处理，忽略全局串行队列设置。适用于并发限制较低的账户。
+                  </p>
+                </div>
+              </label>
+            </div>
+
+            <!-- 拦截预热请求开关（Claude 和 Claude Console） -->
+            <div
+              v-if="form.platform === 'claude' || form.platform === 'claude-console'"
+              class="mt-4"
+            >
+              <label class="flex items-start">
+                <input
+                  v-model="form.interceptWarmup"
+                  class="mt-1 text-blue-600 focus:ring-blue-500 dark:border-gray-600 dark:bg-gray-700"
+                  type="checkbox"
+                />
+                <div class="ml-3">
+                  <span class="text-sm font-medium text-gray-700 dark:text-gray-300">
+                    拦截预热请求
+                  </span>
+                  <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+                    启用后，对标题生成、Warmup 等低价值请求直接返回模拟响应，不消耗上游 API 额度
+                  </p>
+                </div>
+              </label>
+            </div>
+
             <!-- Claude User-Agent 版本配置 -->
             <div v-if="form.platform === 'claude'" class="mt-4">
               <label class="flex items-start">
@@ -1843,7 +1866,7 @@
                     >
                       请从已登录 Gemini CLI 的机器上获取
                       <code class="rounded bg-blue-100 px-1 py-0.5 font-mono dark:bg-blue-900/50"
-                        >~/.config/gemini/credentials.json</code
+                        >~/.config/.gemini/oauth_creds.json</code
                       >
                       文件中的凭证。
                     </p>
@@ -2633,6 +2656,44 @@
             </label>
           </div>
 
+          <!-- Claude 账户级串行队列开关（编辑模式） -->
+          <div v-if="form.platform === 'claude'" class="mt-4">
+            <label class="flex items-start">
+              <input
+                v-model="form.serialQueueEnabled"
+                class="mt-1 text-blue-600 focus:ring-blue-500 dark:border-gray-600 dark:bg-gray-700"
+                type="checkbox"
+              />
+              <div class="ml-3">
+                <span class="text-sm font-medium text-gray-700 dark:text-gray-300">
+                  启用账户级串行队列
+                </span>
+                <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+                  开启后强制该账户的用户消息串行处理，忽略全局串行队列设置。适用于并发限制较低的账户。
+                </p>
+              </div>
+            </label>
+          </div>
+
+          <!-- 拦截预热请求开关（Claude 和 Claude Console 编辑模式） -->
+          <div v-if="form.platform === 'claude' || form.platform === 'claude-console'" class="mt-4">
+            <label class="flex items-start">
+              <input
+                v-model="form.interceptWarmup"
+                class="mt-1 text-blue-600 focus:ring-blue-500 dark:border-gray-600 dark:bg-gray-700"
+                type="checkbox"
+              />
+              <div class="ml-3">
+                <span class="text-sm font-medium text-gray-700 dark:text-gray-300">
+                  拦截预热请求
+                </span>
+                <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+                  启用后，对标题生成、Warmup 等低价值请求直接返回模拟响应，不消耗上游 API 额度
+                </p>
+              </div>
+            </label>
+          </div>
+
           <!-- Claude User-Agent 版本配置（编辑模式） -->
           <div v-if="form.platform === 'claude'" class="mt-4">
             <label class="flex items-start">
@@ -3967,6 +4028,9 @@ const form = ref({
   useUnifiedUserAgent: props.account?.useUnifiedUserAgent || false, // 使用统一Claude Code版本
   useUnifiedClientId: props.account?.useUnifiedClientId || false, // 使用统一的客户端标识
   unifiedClientId: props.account?.unifiedClientId || '', // 统一的客户端标识
+  serialQueueEnabled: (props.account?.maxConcurrency || 0) > 0, // 账户级串行队列开关
+  interceptWarmup:
+    props.account?.interceptWarmup === true || props.account?.interceptWarmup === 'true', // 拦截预热请求
   groupId: '',
   groupIds: [],
   projectId: props.account?.projectId || '',
@@ -4553,9 +4617,11 @@ const buildClaudeAccountData = (tokenInfo, accountName, clientId) => {
     claudeAiOauth: claudeOauthPayload,
     priority: form.value.priority || 50,
     autoStopOnWarning: form.value.autoStopOnWarning || false,
+    interceptWarmup: form.value.interceptWarmup || false,
     useUnifiedUserAgent: form.value.useUnifiedUserAgent || false,
     useUnifiedClientId: form.value.useUnifiedClientId || false,
     unifiedClientId: clientId,
+    maxConcurrency: form.value.serialQueueEnabled ? 1 : 0,
     subscriptionInfo: {
       accountType: form.value.subscriptionType || 'claude_max',
       hasClaudeMax: form.value.subscriptionType === 'claude_max',
@@ -4693,6 +4759,7 @@ const handleOAuthSuccess = async (tokenInfoOrList) => {
       data.useUnifiedUserAgent = form.value.useUnifiedUserAgent || false
       data.useUnifiedClientId = form.value.useUnifiedClientId || false
       data.unifiedClientId = form.value.unifiedClientId || ''
+      data.maxConcurrency = form.value.serialQueueEnabled ? 1 : 0
       // 添加订阅类型信息
       data.subscriptionInfo = {
         accountType: form.value.subscriptionType || 'claude_max',
@@ -5016,6 +5083,7 @@ const createAccount = async () => {
       data.useUnifiedUserAgent = form.value.useUnifiedUserAgent || false
       data.useUnifiedClientId = form.value.useUnifiedClientId || false
       data.unifiedClientId = form.value.unifiedClientId || ''
+      data.maxConcurrency = form.value.serialQueueEnabled ? 1 : 0
       // 添加订阅类型信息
       data.subscriptionInfo = {
         accountType: form.value.subscriptionType || 'claude_max',
@@ -5107,6 +5175,7 @@ const createAccount = async () => {
       // 上游错误处理（仅 Claude Console）
       if (form.value.platform === 'claude-console') {
         data.disableAutoProtection = !!form.value.disableAutoProtection
+        data.interceptWarmup = !!form.value.interceptWarmup
       }
       // 额度管理字段
       data.dailyQuota = form.value.dailyQuota || 0
@@ -5403,9 +5472,11 @@ const updateAccount = async () => {
 
       data.priority = form.value.priority || 50
       data.autoStopOnWarning = form.value.autoStopOnWarning || false
+      data.interceptWarmup = form.value.interceptWarmup || false
       data.useUnifiedUserAgent = form.value.useUnifiedUserAgent || false
       data.useUnifiedClientId = form.value.useUnifiedClientId || false
       data.unifiedClientId = form.value.unifiedClientId || ''
+      data.maxConcurrency = form.value.serialQueueEnabled ? 1 : 0
       // 更新订阅类型信息
       data.subscriptionInfo = {
         accountType: form.value.subscriptionType || 'claude_max',
@@ -5438,6 +5509,8 @@ const updateAccount = async () => {
       data.rateLimitDuration = form.value.enableRateLimit ? form.value.rateLimitDuration || 60 : 0
       // 上游错误处理
       data.disableAutoProtection = !!form.value.disableAutoProtection
+      // 拦截预热请求
+      data.interceptWarmup = !!form.value.interceptWarmup
       // 额度管理字段
       data.dailyQuota = form.value.dailyQuota || 0
       data.quotaResetTime = form.value.quotaResetTime || '00:00'
@@ -6006,9 +6079,12 @@ watch(
         accountType: newAccount.accountType || 'shared',
         subscriptionType: subscriptionType,
         autoStopOnWarning: newAccount.autoStopOnWarning || false,
+        interceptWarmup:
+          newAccount.interceptWarmup === true || newAccount.interceptWarmup === 'true',
         useUnifiedUserAgent: newAccount.useUnifiedUserAgent || false,
         useUnifiedClientId: newAccount.useUnifiedClientId || false,
         unifiedClientId: newAccount.unifiedClientId || '',
+        serialQueueEnabled: (newAccount.maxConcurrency || 0) > 0,
         groupId: groupId,
         groupIds: [],
         projectId: newAccount.projectId || '',

web/admin-spa/src/components/accounts/AccountScheduledTestModal.vue

@@ -0,0 +1,402 @@
+<template>
+  <Teleport to="body">
+    <div
+      v-if="show"
+      class="fixed inset-0 z-[1050] flex items-center justify-center bg-gray-900/40 backdrop-blur-sm"
+    >
+      <div class="absolute inset-0" @click="handleClose" />
+      <div
+        class="relative z-10 mx-3 flex w-full max-w-lg flex-col overflow-hidden rounded-2xl border border-gray-200/70 bg-white/95 shadow-2xl ring-1 ring-black/5 transition-all dark:border-gray-700/60 dark:bg-gray-900/95 dark:ring-white/10 sm:mx-4"
+      >
+        <!-- 顶部栏 -->
+        <div
+          class="flex items-center justify-between border-b border-gray-100 bg-white/80 px-5 py-4 backdrop-blur dark:border-gray-800 dark:bg-gray-900/80"
+        >
+          <div class="flex items-center gap-3">
+            <div
+              class="flex h-10 w-10 flex-shrink-0 items-center justify-center rounded-xl bg-gradient-to-br from-amber-500 to-orange-500 text-white shadow-lg"
+            >
+              <i class="fas fa-clock" />
+            </div>
+            <div>
+              <h3 class="text-lg font-semibold text-gray-900 dark:text-gray-100">定时测试配置</h3>
+              <p class="text-xs text-gray-500 dark:text-gray-400">
+                {{ account?.name || '未知账户' }}
+              </p>
+            </div>
+          </div>
+          <button
+            class="flex h-9 w-9 items-center justify-center rounded-full bg-gray-100 text-gray-500 transition hover:bg-gray-200 hover:text-gray-700 dark:bg-gray-800 dark:text-gray-400 dark:hover:bg-gray-700 dark:hover:text-gray-200"
+            :disabled="saving"
+            @click="handleClose"
+          >
+            <i class="fas fa-times text-sm" />
+          </button>
+        </div>
+
+        <!-- 内容区域 -->
+        <div class="px-5 py-4">
+          <!-- 加载状态 -->
+          <div v-if="loading" class="flex items-center justify-center py-8">
+            <i class="fas fa-spinner fa-spin mr-2 text-blue-500" />
+            <span class="text-gray-500 dark:text-gray-400">加载配置中...</span>
+          </div>
+
+          <template v-else>
+            <!-- 启用开关 -->
+            <div class="mb-5 flex items-center justify-between">
+              <div>
+                <p class="font-medium text-gray-700 dark:text-gray-300">启用定时测试</p>
+                <p class="text-xs text-gray-500 dark:text-gray-400">按计划自动测试账户连通性</p>
+              </div>
+              <button
+                :class="[
+                  'relative h-6 w-11 rounded-full transition-colors duration-200',
+                  config.enabled ? 'bg-green-500' : 'bg-gray-300 dark:bg-gray-600'
+                ]"
+                @click="config.enabled = !config.enabled"
+              >
+                <span
+                  :class="[
+                    'absolute top-0.5 h-5 w-5 rounded-full bg-white shadow-md transition-transform duration-200',
+                    config.enabled ? 'left-5' : 'left-0.5'
+                  ]"
+                />
+              </button>
+            </div>
+
+            <!-- Cron 表达式配置 -->
+            <div class="mb-5">
+              <label class="mb-2 block text-sm font-medium text-gray-700 dark:text-gray-300">
+                Cron 表达式
+              </label>
+              <input
+                v-model="config.cronExpression"
+                class="w-full rounded-lg border border-gray-200 bg-white px-3 py-2 text-sm text-gray-700 placeholder-gray-400 transition focus:border-blue-500 focus:outline-none focus:ring-2 focus:ring-blue-500/20 dark:border-gray-700 dark:bg-gray-800 dark:text-gray-300 dark:placeholder-gray-500"
+                :disabled="!config.enabled"
+                placeholder="0 8 * * *"
+                type="text"
+              />
+              <p class="mt-1.5 text-xs text-gray-500 dark:text-gray-400">
+                格式: 分 时 日 月 周 (例: "0 8 * * *" = 每天8:00)
+              </p>
+            </div>
+
+            <!-- 快捷选项 -->
+            <div class="mb-5">
+              <label class="mb-2 block text-sm font-medium text-gray-700 dark:text-gray-300">
+                快捷设置
+              </label>
+              <div class="flex flex-wrap gap-2">
+                <button
+                  v-for="preset in cronPresets"
+                  :key="preset.value"
+                  :class="[
+                    'rounded-lg border px-3 py-1.5 text-xs font-medium transition',
+                    config.cronExpression === preset.value
+                      ? 'border-blue-500 bg-blue-50 text-blue-700 dark:border-blue-400 dark:bg-blue-900/30 dark:text-blue-300'
+                      : 'border-gray-200 bg-gray-50 text-gray-600 hover:bg-gray-100 dark:border-gray-700 dark:bg-gray-800 dark:text-gray-400 dark:hover:bg-gray-700',
+                    !config.enabled && 'cursor-not-allowed opacity-50'
+                  ]"
+                  :disabled="!config.enabled"
+                  @click="config.cronExpression = preset.value"
+                >
+                  {{ preset.label }}
+                </button>
+              </div>
+            </div>
+
+            <!-- 测试模型选择 -->
+            <div class="mb-5">
+              <label class="mb-2 block text-sm font-medium text-gray-700 dark:text-gray-300">
+                测试模型
+              </label>
+              <input
+                v-model="config.model"
+                class="w-full rounded-lg border border-gray-200 bg-white px-3 py-2 text-sm text-gray-700 placeholder-gray-400 transition focus:border-blue-500 focus:outline-none focus:ring-2 focus:ring-blue-500/20 dark:border-gray-700 dark:bg-gray-800 dark:text-gray-300 dark:placeholder-gray-500"
+                :disabled="!config.enabled"
+                placeholder="claude-sonnet-4-5-20250929"
+                type="text"
+              />
+              <div class="mt-2 flex flex-wrap gap-2">
+                <button
+                  v-for="modelOption in modelOptions"
+                  :key="modelOption.value"
+                  :class="[
+                    'rounded-lg border px-3 py-1.5 text-xs font-medium transition',
+                    config.model === modelOption.value
+                      ? 'border-blue-500 bg-blue-50 text-blue-700 dark:border-blue-400 dark:bg-blue-900/30 dark:text-blue-300'
+                      : 'border-gray-200 bg-gray-50 text-gray-600 hover:bg-gray-100 dark:border-gray-700 dark:bg-gray-800 dark:text-gray-400 dark:hover:bg-gray-700',
+                    !config.enabled && 'cursor-not-allowed opacity-50'
+                  ]"
+                  :disabled="!config.enabled"
+                  @click="config.model = modelOption.value"
+                >
+                  {{ modelOption.label }}
+                </button>
+              </div>
+            </div>
+
+            <!-- 测试历史 -->
+            <div v-if="testHistory.length > 0" class="mb-4">
+              <label class="mb-2 block text-sm font-medium text-gray-700 dark:text-gray-300">
+                最近测试记录
+              </label>
+              <div
+                class="max-h-40 space-y-2 overflow-y-auto rounded-lg border border-gray-200 bg-gray-50 p-3 dark:border-gray-700 dark:bg-gray-800/50"
+              >
+                <div
+                  v-for="(record, index) in testHistory"
+                  :key="index"
+                  class="flex items-center justify-between text-xs"
+                >
+                  <div class="flex items-center gap-2">
+                    <i
+                      :class="[
+                        'fas',
+                        record.success
+                          ? 'fa-check-circle text-green-500'
+                          : 'fa-times-circle text-red-500'
+                      ]"
+                    />
+                    <span class="text-gray-600 dark:text-gray-400">
+                      {{ formatTimestamp(record.timestamp) }}
+                    </span>
+                  </div>
+                  <span v-if="record.latencyMs" class="text-gray-500 dark:text-gray-500">
+                    {{ record.latencyMs }}ms
+                  </span>
+                  <span
+                    v-else-if="record.error"
+                    class="max-w-[150px] truncate text-red-500"
+                    :title="record.error"
+                  >
+                    {{ record.error }}
+                  </span>
+                </div>
+              </div>
+            </div>
+
+            <!-- 无历史记录 -->
+            <div
+              v-else
+              class="mb-4 rounded-lg border border-gray-200 bg-gray-50 p-4 text-center text-sm text-gray-500 dark:border-gray-700 dark:bg-gray-800/50 dark:text-gray-400"
+            >
+              <i class="fas fa-history mb-2 text-2xl text-gray-300 dark:text-gray-600" />
+              <p>暂无测试记录</p>
+            </div>
+          </template>
+        </div>
+
+        <!-- 底部操作栏 -->
+        <div
+          class="flex items-center justify-end gap-3 border-t border-gray-100 bg-gray-50/80 px-5 py-3 dark:border-gray-800 dark:bg-gray-900/50"
+        >
+          <button
+            class="rounded-lg border border-gray-200 bg-white px-4 py-2 text-sm font-medium text-gray-700 shadow-sm transition hover:bg-gray-50 hover:shadow dark:border-gray-700 dark:bg-gray-800 dark:text-gray-300 dark:hover:bg-gray-700"
+            :disabled="saving"
+            @click="handleClose"
+          >
+            取消
+          </button>
+          <button
+            :class="[
+              'flex items-center gap-2 rounded-lg px-4 py-2 text-sm font-medium shadow-sm transition',
+              saving
+                ? 'cursor-not-allowed bg-gray-200 text-gray-400 dark:bg-gray-700 dark:text-gray-500'
+                : 'bg-gradient-to-r from-blue-500 to-indigo-500 text-white hover:from-blue-600 hover:to-indigo-600 hover:shadow-md'
+            ]"
+            :disabled="saving || loading"
+            @click="saveConfig"
+          >
+            <i :class="['fas', saving ? 'fa-spinner fa-spin' : 'fa-save']" />
+            {{ saving ? '保存中...' : '保存配置' }}
+          </button>
+        </div>
+      </div>
+    </div>
+  </Teleport>
+</template>
+
+<script setup>
+import { ref, watch } from 'vue'
+import { API_PREFIX } from '@/config/api'
+import { showToast } from '@/utils/toast'
+
+const props = defineProps({
+  show: {
+    type: Boolean,
+    default: false
+  },
+  account: {
+    type: Object,
+    default: null
+  }
+})
+
+const emit = defineEmits(['close', 'saved'])
+
+// 状态
+const loading = ref(false)
+const saving = ref(false)
+const config = ref({
+  enabled: false,
+  cronExpression: '0 8 * * *',
+  model: 'claude-sonnet-4-5-20250929'
+})
+const testHistory = ref([])
+
+// Cron 预设选项
+const cronPresets = [
+  { label: '每天 8:00', value: '0 8 * * *' },
+  { label: '每天 12:00', value: '0 12 * * *' },
+  { label: '每天 18:00', value: '0 18 * * *' },
+  { label: '每6小时', value: '0 */6 * * *' },
+  { label: '每12小时', value: '0 */12 * * *' },
+  { label: '工作日 9:00', value: '0 9 * * 1-5' }
+]
+
+// 模型选项
+const modelOptions = [
+  { label: 'Claude Sonnet 4.5', value: 'claude-sonnet-4-5-20250929' },
+  { label: 'Claude Haiku 4.5', value: 'claude-haiku-4-5-20251001' },
+  { label: 'Claude Opus 4.5', value: 'claude-opus-4-5-20251101' }
+]
+
+// 格式化时间戳
+function formatTimestamp(timestamp) {
+  if (!timestamp) return '未知'
+  const date = new Date(timestamp)
+  return date.toLocaleString('zh-CN', {
+    month: '2-digit',
+    day: '2-digit',
+    hour: '2-digit',
+    minute: '2-digit'
+  })
+}
+
+// 加载配置
+async function loadConfig() {
+  if (!props.account) return
+
+  loading.value = true
+  try {
+    const authToken = localStorage.getItem('authToken')
+    const platform = props.account.platform
+
+    // 根据平台获取配置端点
+    let endpoint = ''
+    if (platform === 'claude') {
+      endpoint = `${API_PREFIX}/admin/claude-accounts/${props.account.id}/test-config`
+    } else {
+      // 其他平台暂不支持
+      loading.value = false
+      return
+    }
+
+    // 获取配置
+    const configRes = await fetch(endpoint, {
+      headers: {
+        Authorization: authToken ? `Bearer ${authToken}` : ''
+      }
+    })
+
+    if (configRes.ok) {
+      const data = await configRes.json()
+      if (data.success && data.data?.config) {
+        config.value = {
+          enabled: data.data.config.enabled || false,
+          cronExpression: data.data.config.cronExpression || '0 8 * * *',
+          model: data.data.config.model || 'claude-sonnet-4-5-20250929'
+        }
+      }
+    }
+
+    // 获取测试历史
+    const historyEndpoint = endpoint.replace('/test-config', '/test-history')
+    const historyRes = await fetch(historyEndpoint, {
+      headers: {
+        Authorization: authToken ? `Bearer ${authToken}` : ''
+      }
+    })
+
+    if (historyRes.ok) {
+      const historyData = await historyRes.json()
+      if (historyData.success && historyData.data?.history) {
+        testHistory.value = historyData.data.history
+      }
+    }
+  } catch (err) {
+    showToast('加载配置失败: ' + err.message, 'error')
+  } finally {
+    loading.value = false
+  }
+}
+
+// 保存配置
+async function saveConfig() {
+  if (!props.account) return
+
+  saving.value = true
+  try {
+    const authToken = localStorage.getItem('authToken')
+    const platform = props.account.platform
+
+    let endpoint = ''
+    if (platform === 'claude') {
+      endpoint = `${API_PREFIX}/admin/claude-accounts/${props.account.id}/test-config`
+    } else {
+      saving.value = false
+      return
+    }
+
+    const res = await fetch(endpoint, {
+      method: 'PUT',
+      headers: {
+        'Content-Type': 'application/json',
+        Authorization: authToken ? `Bearer ${authToken}` : ''
+      },
+      body: JSON.stringify({
+        enabled: config.value.enabled,
+        cronExpression: config.value.cronExpression,
+        model: config.value.model
+      })
+    })
+
+    if (res.ok) {
+      showToast('配置已保存', 'success')
+      emit('saved')
+      handleClose()
+    } else {
+      const errorData = await res.json().catch(() => ({}))
+      showToast(errorData.message || '保存失败', 'error')
+    }
+  } catch (err) {
+    showToast('保存失败: ' + err.message, 'error')
+  } finally {
+    saving.value = false
+  }
+}
+
+// 关闭模态框
+function handleClose() {
+  if (saving.value) return
+  emit('close')
+}
+
+// 监听 show 变化，加载配置
+watch(
+  () => props.show,
+  (newVal) => {
+    if (newVal) {
+      config.value = {
+        enabled: false,
+        cronExpression: '0 8 * * *',
+        model: 'claude-sonnet-4-5-20250929'
+      }
+      testHistory.value = []
+      loadConfig()
+    }
+  }
+)
+</script>

web/admin-spa/src/components/apistats/LimitConfig.vue

@@ -284,7 +284,7 @@
     </div>
 
     <!-- 详细限制信息 -->
-    <div v-if="hasModelRestrictions" class="card p-4 md:p-6">
+    <div v-if="hasModelRestrictions" class="card !overflow-visible p-4 md:p-6">
       <h3
         class="mb-3 flex items-center text-lg font-bold text-gray-900 dark:text-gray-100 md:mb-4 md:text-xl"
       >
@@ -301,7 +301,7 @@
           <i class="fas fa-robot mr-1 text-xs md:mr-2 md:text-sm" />
           受限模型列表
         </h4>
-        <div class="space-y-1 md:space-y-2">
+        <div class="max-h-64 space-y-1 overflow-y-auto pr-1 md:max-h-80 md:space-y-2">
           <div
             v-for="model in statsData.restrictions.restrictedModels"
             :key="model"

web/admin-spa/src/views/AccountsView.vue

@@ -1238,6 +1238,15 @@
                       <i class="fas fa-vial" />
                       <span class="ml-1">测试</span>
                     </button>
+                    <button
+                      v-if="canTestAccount(account)"
+                      class="rounded bg-amber-100 px-2.5 py-1 text-xs font-medium text-amber-700 transition-colors hover:bg-amber-200 dark:bg-amber-900/40 dark:text-amber-300 dark:hover:bg-amber-800/50"
+                      title="定时测试配置"
+                      @click="openScheduledTestModal(account)"
+                    >
+                      <i class="fas fa-clock" />
+                      <span class="ml-1">定时</span>
+                    </button>
                     <button
                       class="rounded bg-blue-100 px-2.5 py-1 text-xs font-medium text-blue-700 transition-colors hover:bg-blue-200"
                       title="编辑账户"
@@ -1707,6 +1716,15 @@
               测试
             </button>
 
+            <button
+              v-if="canTestAccount(account)"
+              class="flex flex-1 items-center justify-center gap-1 rounded-lg bg-amber-50 px-3 py-2 text-xs text-amber-600 transition-colors hover:bg-amber-100 dark:bg-amber-900/40 dark:text-amber-300 dark:hover:bg-amber-800/50"
+              @click="openScheduledTestModal(account)"
+            >
+              <i class="fas fa-clock" />
+              定时
+            </button>
+
             <button
               class="flex-1 rounded-lg bg-gray-50 px-3 py-2 text-xs text-gray-600 transition-colors hover:bg-gray-100"
               @click="editAccount(account)"
@@ -1880,6 +1898,14 @@
       @close="closeAccountTestModal"
     />
 
+    <!-- 定时测试配置弹窗 -->
+    <AccountScheduledTestModal
+      :account="scheduledTestAccount"
+      :show="showScheduledTestModal"
+      @close="closeScheduledTestModal"
+      @saved="handleScheduledTestSaved"
+    />
+
     <!-- 账户统计弹窗 -->
     <el-dialog
       v-model="showAccountStatsModal"
@@ -2032,6 +2058,7 @@ import CcrAccountForm from '@/components/accounts/CcrAccountForm.vue'
 import AccountUsageDetailModal from '@/components/accounts/AccountUsageDetailModal.vue'
 import AccountExpiryEditModal from '@/components/accounts/AccountExpiryEditModal.vue'
 import AccountTestModal from '@/components/accounts/AccountTestModal.vue'
+import AccountScheduledTestModal from '@/components/accounts/AccountScheduledTestModal.vue'
 import ConfirmModal from '@/components/common/ConfirmModal.vue'
 import CustomDropdown from '@/components/common/CustomDropdown.vue'
 import ActionDropdown from '@/components/common/ActionDropdown.vue'
@@ -2099,6 +2126,10 @@ const expiryEditModalRef = ref(null)
 const showAccountTestModal = ref(false)
 const testingAccount = ref(null)
 
+// 定时测试配置弹窗状态
+const showScheduledTestModal = ref(false)
+const scheduledTestAccount = ref(null)
+
 // 账户统计弹窗状态
 const showAccountStatsModal = ref(false)
 
@@ -2365,6 +2396,13 @@ const getAccountActions = (account) => {
       color: 'blue',
       handler: () => openAccountTestModal(account)
     })
+    actions.push({
+      key: 'scheduled-test',
+      label: '定时测试',
+      icon: 'fa-clock',
+      color: 'amber',
+      handler: () => openScheduledTestModal(account)
+    })
   }
 
   // 删除
@@ -2441,6 +2479,25 @@ const closeAccountTestModal = () => {
   testingAccount.value = null
 }
 
+// 定时测试配置相关函数
+const openScheduledTestModal = (account) => {
+  if (!canTestAccount(account)) {
+    showToast('该账户类型暂不支持定时测试', 'warning')
+    return
+  }
+  scheduledTestAccount.value = account
+  showScheduledTestModal.value = true
+}
+
+const closeScheduledTestModal = () => {
+  showScheduledTestModal.value = false
+  scheduledTestAccount.value = null
+}
+
+const handleScheduledTestSaved = () => {
+  showToast('定时测试配置已保存', 'success')
+}
+
 // 计算排序后的账户列表
 const sortedAccounts = computed(() => {
   let sourceAccounts = accounts.value

web/admin-spa/src/views/SettingsView.vue

@@ -898,6 +898,120 @@
               </div>
             </div>
 
+            <!-- 并发请求排队 -->
+            <div
+              class="mb-6 rounded-lg bg-white/80 p-6 shadow-lg backdrop-blur-sm dark:bg-gray-800/80"
+            >
+              <div class="flex items-center justify-between">
+                <div class="flex items-center">
+                  <div
+                    class="flex h-12 w-12 items-center justify-center rounded-lg bg-gradient-to-r from-blue-500 to-cyan-500 text-white shadow-lg"
+                  >
+                    <i class="fas fa-layer-group text-xl"></i>
+                  </div>
+                  <div class="ml-4">
+                    <h4 class="text-lg font-semibold text-gray-900 dark:text-white">
+                      并发请求排队
+                    </h4>
+                    <p class="text-sm text-gray-500 dark:text-gray-400">
+                      当 API Key 并发请求超限时进入队列等待，而非直接拒绝
+                    </p>
+                  </div>
+                </div>
+                <label class="relative inline-flex cursor-pointer items-center">
+                  <input
+                    v-model="claudeConfig.concurrentRequestQueueEnabled"
+                    class="peer sr-only"
+                    type="checkbox"
+                    @change="saveClaudeConfig"
+                  />
+                  <div
+                    class="peer h-6 w-11 rounded-full bg-gray-200 after:absolute after:left-[2px] after:top-[2px] after:h-5 after:w-5 after:rounded-full after:border after:border-gray-300 after:bg-white after:transition-all after:content-[''] peer-checked:bg-blue-500 peer-checked:after:translate-x-full peer-checked:after:border-white peer-focus:outline-none peer-focus:ring-4 peer-focus:ring-blue-300 dark:border-gray-600 dark:bg-gray-700 dark:peer-focus:ring-blue-800"
+                  ></div>
+                </label>
+              </div>
+
+              <!-- 排队配置详情（仅在启用时显示） -->
+              <div v-if="claudeConfig.concurrentRequestQueueEnabled" class="mt-6 space-y-4">
+                <!-- 固定最小排队数 -->
+                <div>
+                  <label class="block text-sm font-medium text-gray-700 dark:text-gray-300">
+                    <i class="fas fa-list-ol mr-2 text-gray-400"></i>
+                    固定最小排队数
+                  </label>
+                  <input
+                    v-model.number="claudeConfig.concurrentRequestQueueMaxSize"
+                    class="mt-1 block w-full max-w-xs rounded-lg border border-gray-300 bg-white px-3 py-2 shadow-sm focus:border-blue-500 focus:outline-none focus:ring-2 focus:ring-blue-500/20 dark:border-gray-500 dark:bg-gray-700 dark:text-white sm:text-sm"
+                    max="100"
+                    min="1"
+                    placeholder="3"
+                    type="number"
+                    @change="saveClaudeConfig"
+                  />
+                  <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+                    最大排队数的固定最小值（1-100）
+                  </p>
+                </div>
+
+                <!-- 排队数倍数 -->
+                <div>
+                  <label class="block text-sm font-medium text-gray-700 dark:text-gray-300">
+                    <i class="fas fa-times mr-2 text-gray-400"></i>
+                    排队数倍数
+                  </label>
+                  <input
+                    v-model.number="claudeConfig.concurrentRequestQueueMaxSizeMultiplier"
+                    class="mt-1 block w-full max-w-xs rounded-lg border border-gray-300 bg-white px-3 py-2 shadow-sm focus:border-blue-500 focus:outline-none focus:ring-2 focus:ring-blue-500/20 dark:border-gray-500 dark:bg-gray-700 dark:text-white sm:text-sm"
+                    max="10"
+                    min="0"
+                    placeholder="1"
+                    step="0.5"
+                    type="number"
+                    @change="saveClaudeConfig"
+                  />
+                  <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+                    最大排队数 = MAX(倍数 × 并发限制, 固定值)，设为 0 则仅使用固定值
+                  </p>
+                </div>
+
+                <!-- 排队超时时间 -->
+                <div>
+                  <label class="block text-sm font-medium text-gray-700 dark:text-gray-300">
+                    <i class="fas fa-stopwatch mr-2 text-gray-400"></i>
+                    排队超时时间（毫秒）
+                  </label>
+                  <input
+                    v-model.number="claudeConfig.concurrentRequestQueueTimeoutMs"
+                    class="mt-1 block w-full max-w-xs rounded-lg border border-gray-300 bg-white px-3 py-2 shadow-sm focus:border-blue-500 focus:outline-none focus:ring-2 focus:ring-blue-500/20 dark:border-gray-500 dark:bg-gray-700 dark:text-white sm:text-sm"
+                    max="300000"
+                    min="5000"
+                    placeholder="10000"
+                    type="number"
+                    @change="saveClaudeConfig"
+                  />
+                  <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+                    请求在排队中等待的最大时间，超时将返回 429 错误（5秒-5分钟，默认10秒）
+                  </p>
+                </div>
+              </div>
+
+              <div class="mt-4 rounded-lg bg-blue-50 p-4 dark:bg-blue-900/20">
+                <div class="flex">
+                  <i class="fas fa-info-circle mt-0.5 text-blue-500"></i>
+                  <div class="ml-3">
+                    <p class="text-sm text-blue-700 dark:text-blue-300">
+                      <strong>工作原理：</strong>当 API Key 的并发请求超过
+                      <code class="rounded bg-blue-100 px-1 dark:bg-blue-800"
+                        >concurrencyLimit</code
+                      >
+                      时，超限请求会进入队列等待而非直接返回 429。适合 Claude Code Agent
+                      并行工具调用场景。
+                    </p>
+                  </div>
+                </div>
+              </div>
+            </div>
+
             <!-- 配置更新信息 -->
             <div
               v-if="claudeConfig.updatedAt"
@@ -1563,9 +1677,13 @@ const claudeConfig = ref({
   globalSessionBindingEnabled: false,
   sessionBindingErrorMessage: '你的本地session已污染，请清理后使用。',
   sessionBindingTtlDays: 30,
-  userMessageQueueEnabled: true,
+  userMessageQueueEnabled: false, // 与后端默认值保持一致
   userMessageQueueDelayMs: 200,
-  userMessageQueueTimeoutMs: 30000,
+  userMessageQueueTimeoutMs: 5000, // 与后端默认值保持一致（优化后锁持有时间短无需长等待）
+  concurrentRequestQueueEnabled: false,
+  concurrentRequestQueueMaxSize: 3,
+  concurrentRequestQueueMaxSizeMultiplier: 0,
+  concurrentRequestQueueTimeoutMs: 10000,
   updatedAt: null,
   updatedBy: null
 })
@@ -1835,9 +1953,14 @@ const loadClaudeConfig = async () => {
         sessionBindingErrorMessage:
           response.config?.sessionBindingErrorMessage || '你的本地session已污染，请清理后使用。',
         sessionBindingTtlDays: response.config?.sessionBindingTtlDays ?? 30,
-        userMessageQueueEnabled: response.config?.userMessageQueueEnabled ?? true,
+        userMessageQueueEnabled: response.config?.userMessageQueueEnabled ?? false, // 与后端默认值保持一致
         userMessageQueueDelayMs: response.config?.userMessageQueueDelayMs ?? 200,
-        userMessageQueueTimeoutMs: response.config?.userMessageQueueTimeoutMs ?? 30000,
+        userMessageQueueTimeoutMs: response.config?.userMessageQueueTimeoutMs ?? 5000, // 与后端默认值保持一致
+        concurrentRequestQueueEnabled: response.config?.concurrentRequestQueueEnabled ?? false,
+        concurrentRequestQueueMaxSize: response.config?.concurrentRequestQueueMaxSize ?? 3,
+        concurrentRequestQueueMaxSizeMultiplier:
+          response.config?.concurrentRequestQueueMaxSizeMultiplier ?? 0,
+        concurrentRequestQueueTimeoutMs: response.config?.concurrentRequestQueueTimeoutMs ?? 10000,
         updatedAt: response.config?.updatedAt || null,
         updatedBy: response.config?.updatedBy || null
       }
@@ -1865,7 +1988,12 @@ const saveClaudeConfig = async () => {
       sessionBindingTtlDays: claudeConfig.value.sessionBindingTtlDays,
       userMessageQueueEnabled: claudeConfig.value.userMessageQueueEnabled,
       userMessageQueueDelayMs: claudeConfig.value.userMessageQueueDelayMs,
-      userMessageQueueTimeoutMs: claudeConfig.value.userMessageQueueTimeoutMs
+      userMessageQueueTimeoutMs: claudeConfig.value.userMessageQueueTimeoutMs,
+      concurrentRequestQueueEnabled: claudeConfig.value.concurrentRequestQueueEnabled,
+      concurrentRequestQueueMaxSize: claudeConfig.value.concurrentRequestQueueMaxSize,
+      concurrentRequestQueueMaxSizeMultiplier:
+        claudeConfig.value.concurrentRequestQueueMaxSizeMultiplier,
+      concurrentRequestQueueTimeoutMs: claudeConfig.value.concurrentRequestQueueTimeoutMs
     }
 
     const response = await apiClient.put('/admin/claude-relay-config', payload, {