import * as fs from 'fs'; import { getServiceAdapter, getRegisteredProviders } from './adapter.js'; import logger from '../utils/logger.js'; import { MODEL_PROVIDER, getProtocolPrefix } from '../utils/common.js'; import { getProviderModels } from './provider-models.js'; import { broadcastEvent } from '../ui-modules/event-broadcast.js'; import { convertData } from '../convert/convert.js'; import { ENDPOINT_TYPE } from '../utils/common.js'; /** * Manages a pool of API service providers, handling their health and selection. */ export class ProviderPoolManager { // 默认健康检查模型配置 // 键名必须与 MODEL_PROVIDER 常量值一致 static DEFAULT_HEALTH_CHECK_MODELS = { 'gemini-cli-oauth': 'gemini-2.5-flash', 'gemini-antigravity': 'gemini-2.5-flash', 'openai-custom': 'gpt-4o-mini', 'claude-custom': 'claude-3-7-sonnet-20250219', 'claude-kiro-oauth': 'claude-haiku-4-5', 'openai-qwen-oauth': 'qwen3-coder-flash', 'openai-iflow': 'qwen3-coder-plus', 'openai-codex-oauth': 'gpt-5-codex-mini', 'openaiResponses-custom': 'gpt-4o-mini', 'forward-api': 'gpt-4o-mini', }; constructor(providerPools, options = {}) { this.providerPools = providerPools; this.globalConfig = options.globalConfig || {}; // 存储全局配置 this.providerStatus = {}; // Tracks health and usage for each provider instance this.roundRobinIndex = {}; // Tracks the current index for round-robin selection for each provider type // 使用 ?? 运算符确保 0 也能被正确设置，而不是被 || 替换为默认值 this.maxErrorCount = options.maxErrorCount ?? 10; // Default to 10 errors before marking unhealthy this.healthCheckInterval = options.healthCheckInterval ?? 10 * 60 * 1000; // Default to 10 minutes // 日志级别控制 this.logLevel = options.logLevel || 'info'; // 'debug', 'info', 'warn', 'error' // 添加防抖机制，避免频繁的文件 I/O 操作 this.saveDebounceTime = options.saveDebounceTime || 1000; // 默认1秒防抖 this.saveTimer = null; this.pendingSaves = new Set(); // 记录待保存的 providerType // Fallback 链配置 this.fallbackChain = options.globalConfig?.providerFallbackChain || {}; // Model Fallback 映射配置 this.modelFallbackMapping = options.globalConfig?.modelFallbackMapping || {}; // 并发控制：每个 providerType 的选择锁 // 用于确保 selectProvider 的排序 and 更新操作是原子的 this._selectionLocks = {}; this._isSelecting = {}; // 同步标志位锁 // --- V2: 读写分离 and 异步刷新队列 --- // 刷新并发控制配置 this.refreshConcurrency = { global: options.globalConfig?.REFRESH_CONCURRENCY_GLOBAL ?? 2, // 全局最大并行提供商数 perProvider: options.globalConfig?.REFRESH_CONCURRENCY_PER_PROVIDER ?? 1 // 每个提供商内部最大并行数 }; this.activeProviderRefreshes = 0; // 当前正在刷新的提供商类型数量 this.globalRefreshWaiters = []; // 等待全局并发槽位的任务 this.warmupTarget = options.globalConfig?.WARMUP_TARGET || 0; // 默认预热0个节点 this.refreshingUuids = new Set(); // 正在刷新的节点 UUID 集合 this.refreshQueues = {}; // 按 providerType 分组的队列 // 缓冲队列机制：延迟5秒，去重后再执行刷新 this.refreshBufferQueues = {}; // 按 providerType 分组的缓冲队列 this.refreshBufferTimers = {}; // 按 providerType 分组的定时器 this.bufferDelay = options.globalConfig?.REFRESH_BUFFER_DELAY ?? 5000; // 默认5秒缓冲延迟 // 用于并发选点时的原子排序辅助（自增序列） this._selectionSequence = 0; this.initializeProviderStatus(); } /** * 检查所有节点的配置文件，如果发现即将过期则触发刷新 */ async checkAndRefreshExpiringNodes() { this._log('info', 'Checking nodes for approaching expiration dates using provider adapters...'); for (const providerType in this.providerStatus) { const providers = this.providerStatus[providerType]; for (const providerStatus of providers) { const config = providerStatus.config; // 根据 providerType 确定配置文件路径字段名 let configPath = null; if (providerType.startsWith('claude-kiro')) { configPath = config.KIRO_OAUTH_CREDS_FILE_PATH; } else if (providerType.startsWith('gemini-cli')) { configPath = config.GEMINI_OAUTH_CREDS_FILE_PATH; } else if (providerType.startsWith('gemini-antigravity')) { configPath = config.ANTIGRAVITY_OAUTH_CREDS_FILE_PATH; } else if (providerType.startsWith('openai-qwen')) { configPath = config.QWEN_OAUTH_CREDS_FILE_PATH; } else if (providerType.startsWith('openai-iflow')) { configPath = config.IFLOW_OAUTH_CREDS_FILE_PATH; } else if (providerType.startsWith('openai-codex')) { configPath = config.CODEX_OAUTH_CREDS_FILE_PATH; } // logger.info(`Checking node ${providerStatus.uuid} (${providerType}) expiry date... configPath: ${configPath}`); // 排除不健康和禁用的节点 if (!config.isHealthy || config.isDisabled) continue; if (configPath && fs.existsSync(configPath)) { try { const fileContent = fs.readFileSync(configPath, 'utf-8'); const credData = JSON.parse(fileContent); const expiryTime = credData.expiry_date || credData.expiry || credData.expires_at; const nearExpiryMs = (currentConfig?.CRON_NEAR_MINUTES || 10) * 60 * 1000; const isNearExpiry = expiryTime && (expiryTime - Date.now()) < nearExpiryMs; if (isNearExpiry) { this._log('warn', `Node ${providerStatus.uuid} (${providerType}) is near expiration. Enqueuing refresh...`); this._enqueueRefresh(providerType, providerStatus); } } catch (err) { this._log('error', `Failed to check expiry for node ${providerStatus.uuid}: ${err.message}`); } } else { this._log('debug', `Node ${providerStatus.uuid} (${providerType}) has no valid config file path or file does not exist.`); } } } } /** * 系统预热逻辑：按提供商分组，每组预热 warmupTarget 个节点 * @returns {Promise} */ async warmupNodes() { if (this.warmupTarget <= 0) return; this._log('info', `Starting system warmup (Group Target: ${this.warmupTarget} nodes per provider)...`); const nodesToWarmup = []; for (const type in this.providerStatus) { const pool = this.providerStatus[type]; // 挑选当前提供商下需要预热的节点 const candidates = pool .filter(p => p.config.isHealthy && !p.config.isDisabled && !this.refreshingUuids.has(p.uuid)) .sort((a, b) => { // 优先级 A: 明确标记需要刷新的 if (a.config.needsRefresh && !b.config.needsRefresh) return -1; if (!a.config.needsRefresh && b.config.needsRefresh) return 1; // 优先级 B: 按照正常的选择权重排序（最久没用过的优先补） const scoreA = this._calculateNodeScore(a); const scoreB = this._calculateNodeScore(b); return scoreA - scoreB; }) .slice(0, this.warmupTarget); candidates.forEach(p => nodesToWarmup.push({ type, status: p })); } this._log('info', `Warmup: Selected total ${nodesToWarmup.length} nodes across all providers to refresh.`); for (const node of nodesToWarmup) { this._enqueueRefresh(node.type, node.status, true); } // 注意：warmupNodes 不等待队列结束，它是异步后台执行的 } /** * 将节点放入缓冲队列，延迟5秒后去重并执行刷新 * @param {string} providerType * @param {object} providerStatus * @param {boolean} force - 是否强制刷新（跳过缓冲队列） * @private */ _enqueueRefresh(providerType, providerStatus, force = false) { const uuid = providerStatus.uuid; // 如果已经在刷新中，直接返回 if (this.refreshingUuids.has(uuid)) { this._log('debug', `Node ${uuid} is already in refresh queue.`); return; } // 判断提供商池内的总可用节点数，小于5个时，不等待缓冲，直接加入刷新队列 const healthyCount = this.getHealthyCount(providerType); if (healthyCount < 5) { this._log('info', `Provider ${providerType} has only ${healthyCount} healthy nodes. Bypassing buffer and enqueuing refresh for ${uuid} immediately.`); this._enqueueRefreshImmediate(providerType, providerStatus, force); return; } // 初始化缓冲队列 if (!this.refreshBufferQueues[providerType]) { this.refreshBufferQueues[providerType] = new Map(); // 使用 Map 自动去重 } const bufferQueue = this.refreshBufferQueues[providerType]; // 检查是否已在缓冲队列中 const existing = bufferQueue.get(uuid); const isNewEntry = !existing; // 更新或添加节点（保留 force: true 状态） bufferQueue.set(uuid, { providerStatus, force: existing ? (existing.force || force) : force }); if (isNewEntry) { this._log('debug', `Node ${uuid} added to buffer queue for ${providerType}. Buffer size: ${bufferQueue.size}`); } else { this._log('debug', `Node ${uuid} already in buffer queue, updated force flag. Buffer size: ${bufferQueue.size}`); } // 只在新增节点或缓冲队列为空时重置定时器 // 避免频繁重置导致刷新被无限延迟 if (isNewEntry || !this.refreshBufferTimers[providerType]) { // 清除之前的定时器 if (this.refreshBufferTimers[providerType]) { clearTimeout(this.refreshBufferTimers[providerType]); } // 设置新的定时器，延迟5秒后处理缓冲队列 this.refreshBufferTimers[providerType] = setTimeout(() => { this._flushRefreshBuffer(providerType); }, this.bufferDelay); } } /** * 处理缓冲队列，将去重后的节点放入实际刷新队列 * @param {string} providerType * @private */ _flushRefreshBuffer(providerType) { const bufferQueue = this.refreshBufferQueues[providerType]; if (!bufferQueue || bufferQueue.size === 0) { return; } this._log('info', `Flushing refresh buffer for ${providerType}. Processing ${bufferQueue.size} unique nodes.`); // 将缓冲队列中的所有节点放入实际刷新队列 for (const [uuid, { providerStatus, force }] of bufferQueue.entries()) { this._enqueueRefreshImmediate(providerType, providerStatus, force); } // 清空缓冲队列和定时器 bufferQueue.clear(); delete this.refreshBufferTimers[providerType]; } /** * 立即将节点放入刷新队列（内部方法，由缓冲队列调用） * @param {string} providerType * @param {object} providerStatus * @param {boolean} force * @private */ _enqueueRefreshImmediate(providerType, providerStatus, force = false) { const uuid = providerStatus.uuid; // 再次检查是否已经在刷新中（防止并发问题） if (this.refreshingUuids.has(uuid)) { this._log('debug', `Node ${uuid} is already in refresh queue (immediate check).`); return; } this.refreshingUuids.add(uuid); // 初始化提供商队列 if (!this.refreshQueues[providerType]) { this.refreshQueues[providerType] = { activeCount: 0, waitingTasks: [] }; } const queue = this.refreshQueues[providerType]; // 记录此任务是否持有一个全局槽位（情况1追加的任务不持有） let ownsGlobalSlot = false; const runTask = async () => { try { await this._refreshNodeToken(providerType, providerStatus, force); } catch (err) { this._log('error', `Failed to process refresh for node ${uuid}: ${err.message}`); } finally { this.refreshingUuids.delete(uuid); // 再次获取当前队列引用 const currentQueue = this.refreshQueues[providerType]; if (!currentQueue) return; currentQueue.activeCount--; // 1. 尝试从当前提供商队列中取下一个任务 if (currentQueue.waitingTasks.length > 0) { const nextTask = currentQueue.waitingTasks.shift(); currentQueue.activeCount++; // 使用 Promise.resolve().then 避免过深的递归 Promise.resolve().then(nextTask); } else if (currentQueue.activeCount === 0) { // 2. 如果当前提供商的所有任务都完成了，释放全局槽位 // 只有持有全局槽位的任务才能递减计数器，避免负值 if (ownsGlobalSlot && currentQueue.waitingTasks.length === 0 && this.refreshQueues[providerType] === currentQueue) { this.activeProviderRefreshes--; delete this.refreshQueues[providerType]; // 清理空队列 } // 3. 尝试启动下一个等待中的提供商队列 if (this.globalRefreshWaiters.length > 0) { const nextProviderStart = this.globalRefreshWaiters.shift(); Promise.resolve().then(nextProviderStart); } } } }; const tryStartProviderQueue = () => { if (queue.activeCount < this.refreshConcurrency.perProvider) { queue.activeCount++; runTask(); } else { queue.waitingTasks.push(runTask); } }; // 检查全局并发限制（按提供商分组） // 情况1: 该提供商已经在运行，直接加入其队列（不占用新的全局槽位） const isExistingQueue = this.refreshQueues[providerType].activeCount > 0 || this.refreshQueues[providerType].waitingTasks.length > 0; if (isExistingQueue) { tryStartProviderQueue(); } // 情况2: 该提供商未运行，需要检查全局槽位，此路径持有全局槽位 else if (this.activeProviderRefreshes < this.refreshConcurrency.global) { ownsGlobalSlot = true; this.activeProviderRefreshes++; tryStartProviderQueue(); } // 情况3: 全局槽位已满，进入等待队列，由等待回调负责标记持槽 else { this.globalRefreshWaiters.push(() => { // 重新获取最新的队列引用 if (!this.refreshQueues[providerType]) { this.refreshQueues[providerType] = { activeCount: 0, waitingTasks: [] }; } // 从等待队列启动时持有全局槽位 ownsGlobalSlot = true; this.activeProviderRefreshes++; tryStartProviderQueue(); }); } } /** * 实际执行节点刷新逻辑 * @private */ async _refreshNodeToken(providerType, providerStatus, force = false) { const config = providerStatus.config; // 检查刷新次数是否已达上限（最大5次） const currentRefreshCount = config.refreshCount || 0; if (currentRefreshCount >= 5 && !force) { this._log('warn', `Node ${providerStatus.uuid} has reached maximum refresh count (5), marking as unhealthy`); // 标记为不健康 this.markProviderUnhealthyImmediately(providerType, config, 'Maximum refresh count (5) reached'); return; } // 添加5秒内的随机等待时间，避免并发刷新时的冲突 // const randomDelay = Math.floor(Math.random() * 5000); // this._log('info', `Starting token refresh for node ${providerStatus.uuid} (${providerType}) with ${randomDelay}ms delay`); // await new Promise(resolve => setTimeout(resolve, randomDelay)); try { // 增加刷新计数 config.refreshCount = currentRefreshCount + 1; // 使用适配器进行刷新 const tempConfig = { ...this.globalConfig, ...config, MODEL_PROVIDER: providerType }; const serviceAdapter = getServiceAdapter(tempConfig); // 调用适配器的 refreshToken 方法（内部封装了具体的刷新逻辑） if (typeof serviceAdapter.refreshToken === 'function') { const startTime = Date.now(); if (force) { if (typeof serviceAdapter.forceRefreshToken === 'function') { await serviceAdapter.forceRefreshToken(); } else { this._log('warn', `forceRefreshToken not implemented for ${providerType}, falling back to refreshToken`); await serviceAdapter.refreshToken(); } } else { await serviceAdapter.refreshToken(); } const duration = Date.now() - startTime; this._log('info', `Token refresh successful for node ${providerStatus.uuid} (Duration: ${duration}ms)`); // 刷新成功，统一重置状态 config.needsRefresh = false; config.refreshCount = 0; config.lastRefreshTime = Date.now(); // 记录最后刷新成功时间 } else { throw new Error(`refreshToken method not implemented for ${providerType}`); } } catch (error) { this._log('error', `Token refresh failed for node ${providerStatus.uuid}: ${error.message}`); this.markProviderUnhealthyImmediately(providerType, config, `Refresh failed: ${error.message}`); throw error; } } /** * 计算节点的权重/评分，用于排序 * 分数越低，优先级越高 * @private */ _calculateNodeScore(providerStatus, now = Date.now(), minSeqInPool = -1) { const config = providerStatus.config; const state = providerStatus.state; // 1. 基础健康分：不健康的排最后 if (!config.isHealthy || config.isDisabled) return 1e18; // 检查并发限制 const concurrencyLimit = parseInt(config.concurrencyLimit || 0); const queueLimit = parseInt(config.queueLimit || 0); if (concurrencyLimit > 0) { if (state.activeCount >= concurrencyLimit) { // 如果队列也满了，排在最后（但优于不健康节点） if (queueLimit > 0 && state.waitingCount >= queueLimit) { return 1e17; } // 没满，但需要排队。排队数量越多，权重越大 return 1e15 + (state.waitingCount || 0) * 1e10; } } // 2. 预热/新鲜度判断 const lastHealthCheckTime = config.lastHealthCheckTime ? new Date(config.lastHealthCheckTime).getTime() : 0; const isFresh = lastHealthCheckTime && (now - lastHealthCheckTime < 60000); // 3. 计算统一评分 // 基础分：新鲜节点使用固定负偏移 (-1e14)，普通节点使用上次使用时间 (约 1.7e12) const lastUsedTime = config.lastUsed ? new Date(config.lastUsed).getTime() : (now - 86400000); const baseScore = isFresh ? -1e14 : lastUsedTime; // 惩罚项 A: 使用次数 (每多用一次增加 10 秒权重) const usageCount = config.usageCount || 0; const usageScore = usageCount * 10000; // 惩罚项 B: 相对序列号 (用于打破平局，确保轮询) const lastSelectionSeq = config._lastSelectionSeq || 0; if (minSeqInPool === -1) { const pool = this.providerStatus[providerStatus.type] || []; minSeqInPool = pool.reduce((min, p) => Math.min(min, p.config._lastSelectionSeq || 0), Infinity); } const relativeSeq = Math.max(0, lastSelectionSeq - minSeqInPool); const cappedRelativeSeq = Math.min(relativeSeq, 100); const sequenceScore = cappedRelativeSeq * 1000; // 惩罚项 C: 负载 (每个活跃请求增加 5 秒权重) const loadScore = (state.activeCount || 0) * 5000; // 新鲜节点的微调：配合 usageScore 和 sequenceScore 在多个新鲜节点间轮询 const freshBonus = isFresh ? (now - lastHealthCheckTime) : 0; return baseScore + usageScore + sequenceScore + loadScore + freshBonus; } /** * 获取指定类型的健康节点数量 */ getHealthyCount(providerType) { return (this.providerStatus[providerType] || []).filter(p => p.config.isHealthy && !p.config.isDisabled).length; } /** * 日志输出方法，支持日志级别控制 * @private */ _log(level, message) { const levels = { debug: 0, info: 1, warn: 2, error: 3 }; if (levels[level] >= levels[this.logLevel]) { logger[level](`[ProviderPoolManager] ${message}`); } } /** * 记录健康状态变化日志 * @param {string} providerType - 提供商类型 * @param {object} providerConfig - 提供商配置 * @param {string} fromStatus - 之前状态 * @param {string} toStatus - 当前状态 * @param {string} [errorMessage] - 错误信息（可选） * @private */ _logHealthStatusChange(providerType, providerConfig, fromStatus, toStatus, errorMessage = null) { const customName = providerConfig.customName || providerConfig.uuid; const timestamp = new Date().toISOString(); const logEntry = { timestamp, providerType, uuid: providerConfig.uuid, customName, fromStatus, toStatus, errorMessage, usageCount: providerConfig.usageCount || 0, errorCount: providerConfig.errorCount || 0 }; // 输出详细的状态变化日志 if (toStatus === 'unhealthy') { logger.warn(`[HealthMonitor] ⚠️ Provider became UNHEALTHY: ${customName} (${providerType})`); logger.warn(`[HealthMonitor] Reason: ${errorMessage || 'Unknown'}`); logger.warn(`[HealthMonitor] Error Count: ${providerConfig.errorCount}`); // 触发告警（如果配置了 Webhook） this._triggerHealthAlert(providerType, providerConfig, 'unhealthy', errorMessage); } else if (toStatus === 'healthy' && fromStatus === 'unhealthy') { logger.info(`[HealthMonitor] ✅ Provider recovered to HEALTHY: ${customName} (${providerType})`); // 触发恢复通知 this._triggerHealthAlert(providerType, providerConfig, 'recovered', null); } // 广播健康状态变化事件 broadcastEvent('health_status_change', logEntry); } /** * 触发健康状态告警 * @param {string} providerType - 提供商类型 * @param {object} providerConfig - 提供商配置 * @param {string} status - 状态 ('unhealthy' | 'recovered') * @param {string} [errorMessage] - 错误信息 * @private */ async _triggerHealthAlert(providerType, providerConfig, status, errorMessage = null) { const webhookUrl = this.globalConfig?.HEALTH_ALERT_WEBHOOK_URL; if (!webhookUrl) { return; // 未配置 Webhook，跳过 } const customName = providerConfig.customName || providerConfig.uuid; const payload = { timestamp: new Date().toISOString(), providerType, uuid: providerConfig.uuid, customName, status, errorMessage, stats: { usageCount: providerConfig.usageCount || 0, errorCount: providerConfig.errorCount || 0 } }; try { const axios = (await import('axios')).default; await axios.post(webhookUrl, payload, { timeout: 5000, headers: { 'Content-Type': 'application/json' } }); this._log('info', `Health alert sent to webhook for ${customName}: ${status}`); } catch (error) { this._log('error', `Failed to send health alert to webhook: ${error.message}`); } } /** * 查找指定的 provider * @private */ _findProvider(providerType, uuid) { if (!providerType || !uuid) { this._log('error', `Invalid parameters: providerType=${providerType}, uuid=${uuid}`); return null; } const pool = this.providerStatus[providerType]; return pool?.find(p => p.uuid === uuid) || null; } /** * 根据 UUID 在所有池中查找提供商配置 * @param {string} uuid - 提供商 UUID * @returns {object|null} 提供商配置对象或 null */ findProviderByUuid(uuid) { if (!uuid) return null; for (const type in this.providerStatus) { const provider = this.providerStatus[type].find(p => p.uuid === uuid); if (provider) return provider.config; } return null; } /** * Initializes the status for each provider in the pools. * Initially, all providers are considered healthy and have zero usage. */ initializeProviderStatus() { for (const providerType in this.providerPools) { const oldStatus = this.providerStatus[providerType] || []; this.providerStatus[providerType] = []; this.roundRobinIndex[providerType] = 0; // Initialize round-robin index for each type // 只有在锁不存在时才初始化，避免在运行中被重置导致并发问题 if (!this._selectionLocks[providerType]) { this._selectionLocks[providerType] = Promise.resolve(); } this.providerPools[providerType].forEach((providerConfig) => { // 尝试从旧状态中恢复活跃请求计数和队列，避免重载配置时重置并发限制 const existing = oldStatus.find(p => p.uuid === providerConfig.uuid); // Ensure initial health and usage stats are present in the config providerConfig.isHealthy = providerConfig.isHealthy !== undefined ? providerConfig.isHealthy : true; providerConfig.isDisabled = providerConfig.isDisabled !== undefined ? providerConfig.isDisabled : false; providerConfig.lastUsed = providerConfig.lastUsed !== undefined ? providerConfig.lastUsed : null; providerConfig.usageCount = providerConfig.usageCount !== undefined ? providerConfig.usageCount : 0; providerConfig.errorCount = providerConfig.errorCount !== undefined ? providerConfig.errorCount : 0; // --- V2: 刷新监控字段 --- providerConfig.needsRefresh = providerConfig.needsRefresh !== undefined ? providerConfig.needsRefresh : false; providerConfig.refreshCount = providerConfig.refreshCount !== undefined ? providerConfig.refreshCount : 0; // 优化2: 简化 lastErrorTime 处理逻辑 providerConfig.lastErrorTime = providerConfig.lastErrorTime instanceof Date ? providerConfig.lastErrorTime.toISOString() : (providerConfig.lastErrorTime || null); // 健康检测相关字段 providerConfig.lastHealthCheckTime = providerConfig.lastHealthCheckTime || null; providerConfig.lastHealthCheckModel = providerConfig.lastHealthCheckModel || null; providerConfig.lastErrorMessage = providerConfig.lastErrorMessage || null; providerConfig.customName = providerConfig.customName || null; this.providerStatus[providerType].push({ config: providerConfig, uuid: providerConfig.uuid, // Still keep uuid at the top level for easy access type: providerType, // 保存 providerType 引用 state: existing ? existing.state : { activeCount: 0, waitingCount: 0, queue: [] } }); }); } this._log('info', `Initialized provider statuses: ok (maxErrorCount: ${this.maxErrorCount})`); } /** * 获取一个可用的提供商插槽，考虑并发限制和队列 * @param {string} providerType * @param {string} requestedModel * @param {object} options */ async acquireSlot(providerType, requestedModel = null, options = {}) { // 使用 selectProvider 进行初次选择（评分逻辑已经包含了并发考虑） const selectedConfig = await this.selectProvider(providerType, requestedModel, { ...options, skipUsageCount: true }); if (!selectedConfig) { return null; } const provider = this._findProvider(providerType, selectedConfig.uuid); if (!provider) return selectedConfig; const config = provider.config; const state = provider.state; const concurrencyLimit = parseInt(config.concurrencyLimit || 0); const queueLimit = parseInt(config.queueLimit || 0); // 如果没有限制，直接增加活跃计数并返回 if (concurrencyLimit <= 0) { state.activeCount++; return config; } // 检查是否在并发限制内 if (state.activeCount < concurrencyLimit) { state.activeCount++; return config; } // 超过并发限制，尝试进入队列 if (queueLimit > 0 && state.waitingCount < queueLimit) { this._log('info', `[Concurrency] Node ${config.uuid} busy (${state.activeCount}/${concurrencyLimit}), enqueuing request (queue: ${state.waitingCount + 1}/${queueLimit})`); state.waitingCount++; try { // 等待释放信号 await new Promise((resolve, reject) => { // 设置较短的超时用于测试验证，或者由外部控制 const timeoutMs = options.queueTimeout || 300000; const timeout = setTimeout(() => { const idx = state.queue.indexOf(handler); if (idx !== -1) { state.queue.splice(idx, 1); reject(new Error(`Queue timeout after ${timeoutMs/1000}s`)); } }, timeoutMs); const handler = () => { clearTimeout(timeout); resolve(); }; state.queue.push(handler); }); } finally { state.waitingCount--; } // 获得信号后，增加活跃计数 state.activeCount++; return config; } // 队列也满了 this._log('warn', `[Concurrency] Node ${config.uuid} full capacity (${state.activeCount}/${concurrencyLimit}, queue: ${state.waitingCount}/${queueLimit}), returning 429`); const error = new Error('Too many requests: account concurrency limit and queue reached'); error.status = 429; error.code = 429; throw error; } /** * 释放提供商插槽 */ releaseSlot(providerType, uuid) { if (!providerType || !uuid) return; const provider = this._findProvider(providerType, uuid); if (!provider) return; const state = provider.state; if (state.activeCount > 0) { state.activeCount--; } // 如果队列中有等待的任务，释放下一个 if (state.queue && state.queue.length > 0) { const next = state.queue.shift(); if (next) { // 异步触发 setImmediate(next); } } } /** * Selects a provider from the pool for a given provider type. * Currently uses a simple round-robin for healthy providers. * If requestedModel is provided, providers that don't support the model will be excluded. * * 注意：此方法现在返回 Promise，使用互斥锁确保并发安全。 * * @param {string} providerType - The type of provider to select (e.g., 'gemini-cli', 'openai-custom'). * @param {string} [requestedModel] - Optional. The model name to filter providers by. * @returns {Promise} The selected provider's configuration, or null if no healthy provider is found. */ async selectProvider(providerType, requestedModel = null, options = {}) { // 参数校验 if (!providerType || typeof providerType !== 'string') { this._log('error', `Invalid providerType: ${providerType}`); return null; } // 使用标志位 + 异步等待实现更强力的互斥锁 // 这种方式能更好地处理同一微任务循环内的并发 while (this._isSelecting[providerType]) { await new Promise(resolve => setImmediate(resolve)); } this._isSelecting[providerType] = true; try { // 在锁内部执行同步选择 return this._doSelectProvider(providerType, requestedModel, options); } finally { this._isSelecting[providerType] = false; } } /** * 实际执行 provider 选择的内部方法（同步执行，由锁保护） * @private */ _doSelectProvider(providerType, requestedModel, options) { const availableProviders = this.providerStatus[providerType] || []; // 检查并恢复已到恢复时间的提供商 this._checkAndRecoverScheduledProviders(providerType); // 获取固定时间戳，确保排序过程中一致 const now = Date.now(); // 提前计算池中最小序列号，避免在排序算法中重复 O(N) 计算 const minSeq = Math.min(...availableProviders.map(p => p.config._lastSelectionSeq || 0)); let availableAndHealthyProviders = availableProviders.filter(p => p.config.isHealthy && !p.config.isDisabled && !p.config.needsRefresh ); // 如果指定了模型，则排除不支持该模型的提供商 if (requestedModel) { const modelFilteredProviders = availableAndHealthyProviders.filter(p => { // 如果提供商没有配置 notSupportedModels，则认为它支持所有模型 if (!p.config.notSupportedModels || !Array.isArray(p.config.notSupportedModels)) { return true; } // 检查 notSupportedModels 数组中是否包含请求的模型，如果包含则排除 return !p.config.notSupportedModels.includes(requestedModel); }); if (modelFilteredProviders.length === 0) { this._log('warn', `No available providers for type: ${providerType} that support model: ${requestedModel}`); return null; } availableAndHealthyProviders = modelFilteredProviders; this._log('debug', `Filtered ${modelFilteredProviders.length} providers supporting model: ${requestedModel}`); } if (availableAndHealthyProviders.length === 0) { this._log('warn', `No available and healthy providers for type: ${providerType}`); return null; } // 改进：使用统一的评分策略进行选择 // 传入当前时间戳 now 确保一致性 const selected = availableAndHealthyProviders.sort((a, b) => { const scoreA = this._calculateNodeScore(a, now, minSeq); const scoreB = this._calculateNodeScore(b, now, minSeq); if (scoreA !== scoreB) return scoreA - scoreB; // 如果分值相同，使用 UUID 排序确保确定性 return a.uuid < b.uuid ? -1 : 1; })[0]; // 始终更新 lastUsed（确保 LRU 策略生效，避免并发请求选到同一个 provider） // usageCount 只在请求成功后才增加（由 skipUsageCount 控制） selected.config.lastUsed = new Date().toISOString(); // 更新自增序列号，确保即使毫秒级并发，也能在下一轮排序中被区分开 this._selectionSequence++; selected.config._lastSelectionSeq = this._selectionSequence; // 强制打印选中日志，方便排查并发问题 this._log('info', `[Concurrency Control] Atomic selection: ${selected.config.uuid} (Seq: ${this._selectionSequence})`); if (!options.skipUsageCount) { selected.config.usageCount++; } // 使用防抖保存（文件 I/O 是异步的，但内存已经更新） this._debouncedSave(providerType); this._log('debug', `Selected provider for ${providerType} (LRU): ${selected.config.uuid}${requestedModel ? ` for model: ${requestedModel}` : ''}${options.skipUsageCount ? ' (skip usage count)' : ''}`); return selected.config; } /** * 获取一个可用的提供商插槽，支持 Fallback 机制 */ async acquireSlotWithFallback(providerType, requestedModel = null, options = {}) { if (!providerType || typeof providerType !== 'string') { this._log('error', `Invalid providerType: ${providerType}`); return null; } const triedTypes = new Set(); const typesToTry = [providerType]; const fallbackTypes = this.fallbackChain[providerType] || []; if (Array.isArray(fallbackTypes)) { typesToTry.push(...fallbackTypes); } for (const currentType of typesToTry) { if (triedTypes.has(currentType)) continue; triedTypes.add(currentType); if (!this.providerStatus[currentType] || this.providerStatus[currentType].length === 0) { continue; } if (currentType !== providerType && requestedModel) { const primaryProtocol = getProtocolPrefix(providerType); const fallbackProtocol = getProtocolPrefix(currentType); if (primaryProtocol !== fallbackProtocol) continue; const supportedModels = getProviderModels(currentType); if (supportedModels.length > 0 && !supportedModels.includes(requestedModel)) continue; } // 尝试获取插槽 try { const selectedConfig = await this.acquireSlot(currentType, requestedModel, options); if (selectedConfig) { if (currentType !== providerType) { this._log('info', `Fallback Slot activated (Chain): ${providerType} -> ${currentType} (uuid: ${selectedConfig.uuid})`); } return { config: selectedConfig, actualProviderType: currentType, isFallback: currentType !== providerType }; } } catch (err) { if (err.status === 429) { // 如果是因为 429 (并发/队列满)，尝试下一个 Fallback this._log('info', `Type ${currentType} busy (429), trying next fallback...`); continue; } throw err; // 其他错误抛出 } } // Model Fallback Mapping if (requestedModel && this.modelFallbackMapping && this.modelFallbackMapping[requestedModel]) { const mapping = this.modelFallbackMapping[requestedModel]; const targetProviderType = mapping.targetProviderType; const targetModel = mapping.targetModel; if (targetProviderType && targetModel) { if (this.providerStatus[targetProviderType] && this.providerStatus[targetProviderType].length > 0) { try { const selectedConfig = await this.acquireSlot(targetProviderType, targetModel, options); if (selectedConfig) { return { config: selectedConfig, actualProviderType: targetProviderType, isFallback: true, actualModel: targetModel }; } } catch (err) { // 如果目标类型繁忙，尝试它的 fallback chain const targetFallbackTypes = this.fallbackChain[targetProviderType] || []; for (const fallbackType of targetFallbackTypes) { const targetProtocol = getProtocolPrefix(targetProviderType); const fallbackProtocol = getProtocolPrefix(fallbackType); if (targetProtocol !== fallbackProtocol) continue; const supportedModels = getProviderModels(fallbackType); if (supportedModels.length > 0 && !supportedModels.includes(targetModel)) continue; try { const fallbackSelectedConfig = await this.acquireSlot(fallbackType, targetModel, options); if (fallbackSelectedConfig) { return { config: fallbackSelectedConfig, actualProviderType: fallbackType, isFallback: true, actualModel: targetModel }; } } catch (e) { continue; } } } } } } return null; } /** * Selects a provider from the pool with fallback support. * When the primary provider type has no healthy providers, it will try fallback types. * @param {string} providerType - The primary type of provider to select. * @param {string} [requestedModel] - Optional. The model name to filter providers by. * @param {Object} [options] - Optional. Additional options. * @param {boolean} [options.skipUsageCount] - Optional. If true, skip incrementing usage count. * @returns {object|null} An object containing the selected provider's configuration and the actual provider type used, or null if no healthy provider is found. */ /** * Selects a provider from the pool with fallback support. * When the primary provider type has no healthy providers, it will try fallback types. * * 注意：此方法现在返回 Promise，因为内部调用的 selectProvider 是异步的。 * * @param {string} providerType - The primary type of provider to select. * @param {string} [requestedModel] - Optional. The model name to filter providers by. * @param {Object} [options] - Optional. Additional options. * @param {boolean} [options.skipUsageCount] - Optional. If true, skip incrementing usage count. * @returns {Promise} An object containing the selected provider's configuration and the actual provider type used, or null if no healthy provider is found. */ async selectProviderWithFallback(providerType, requestedModel = null, options = {}) { // 参数校验 if (!providerType || typeof providerType !== 'string') { this._log('error', `Invalid providerType: ${providerType}`); return null; } // ========================== // 优先级 1: Provider Fallback Chain (同协议/兼容协议的回退) // ========================== // 记录尝试过的类型，避免循环 const triedTypes = new Set(); const typesToTry = [providerType]; const fallbackTypes = this.fallbackChain[providerType] || []; if (Array.isArray(fallbackTypes)) { typesToTry.push(...fallbackTypes); } for (const currentType of typesToTry) { // 避免重复尝试 if (triedTypes.has(currentType)) { continue; } triedTypes.add(currentType); // 检查该类型是否有配置的池 if (!this.providerStatus[currentType] || this.providerStatus[currentType].length === 0) { this._log('debug', `No provider pool configured for type: ${currentType}`); continue; } // 如果是 fallback 类型，需要检查模型兼容性 if (currentType !== providerType && requestedModel) { // 检查协议前缀是否兼容 const primaryProtocol = getProtocolPrefix(providerType); const fallbackProtocol = getProtocolPrefix(currentType); if (primaryProtocol !== fallbackProtocol) { this._log('debug', `Skipping fallback type ${currentType}: protocol mismatch (${primaryProtocol} vs ${fallbackProtocol})`); continue; } // 检查 fallback 类型是否支持请求的模型 const supportedModels = getProviderModels(currentType); if (supportedModels.length > 0 && !supportedModels.includes(requestedModel)) { this._log('debug', `Skipping fallback type ${currentType}: model ${requestedModel} not supported`); continue; } } // 尝试从当前类型选择提供商（现在是异步的） const selectedConfig = await this.selectProvider(currentType, requestedModel, options); if (selectedConfig) { if (currentType !== providerType) { this._log('info', `Fallback activated (Chain): ${providerType} -> ${currentType} (uuid: ${selectedConfig.uuid})`); } return { config: selectedConfig, actualProviderType: currentType, isFallback: currentType !== providerType }; } } // ========================== // 优先级 2: Model Fallback Mapping (跨协议/特定模型的回退) // ========================== if (requestedModel && this.modelFallbackMapping && this.modelFallbackMapping[requestedModel]) { const mapping = this.modelFallbackMapping[requestedModel]; const targetProviderType = mapping.targetProviderType; const targetModel = mapping.targetModel; if (targetProviderType && targetModel) { this._log('info', `Trying Model Fallback Mapping for ${requestedModel}: -> ${targetProviderType} (${targetModel})`); // 递归调用 selectProviderWithFallback，但这次针对目标提供商类型 // 注意：这里我们直接尝试从目标提供商池中选择，因为如果再次递归可能会导致死循环或逻辑复杂化 // 简单起见，我们直接尝试选择目标提供商 // 检查目标类型是否有配置的池 if (this.providerStatus[targetProviderType] && this.providerStatus[targetProviderType].length > 0) { // 尝试从目标类型选择提供商（使用转换后的模型名，现在是异步的） const selectedConfig = await this.selectProvider(targetProviderType, targetModel, options); if (selectedConfig) { this._log('info', `Fallback activated (Model Mapping): ${providerType} (${requestedModel}) -> ${targetProviderType} (${targetModel}) (uuid: ${selectedConfig.uuid})`); return { config: selectedConfig, actualProviderType: targetProviderType, isFallback: true, actualModel: targetModel // 返回实际使用的模型名，供上层进行请求转换 }; } else { // 如果目标类型的主池也不可用，尝试目标类型的 fallback chain // 例如 claude-kiro-oauth (mapped) -> claude-custom (chain) // 这需要我们小心处理，避免无限递归。 // 我们可以手动检查目标类型的 fallback chain const targetFallbackTypes = this.fallbackChain[targetProviderType] || []; for (const fallbackType of targetFallbackTypes) { // 检查协议兼容性 (目标类型 vs 它的 fallback) const targetProtocol = getProtocolPrefix(targetProviderType); const fallbackProtocol = getProtocolPrefix(fallbackType); if (targetProtocol !== fallbackProtocol) continue; // 检查模型支持 const supportedModels = getProviderModels(fallbackType); if (supportedModels.length > 0 && !supportedModels.includes(targetModel)) continue; const fallbackSelectedConfig = await this.selectProvider(fallbackType, targetModel, options); if (fallbackSelectedConfig) { this._log('info', `Fallback activated (Model Mapping -> Chain): ${providerType} (${requestedModel}) -> ${targetProviderType} -> ${fallbackType} (${targetModel}) (uuid: ${fallbackSelectedConfig.uuid})`); return { config: fallbackSelectedConfig, actualProviderType: fallbackType, isFallback: true, actualModel: targetModel }; } } } } else { this._log('warn', `Model Fallback target provider ${targetProviderType} not configured or empty.`); } } } this._log('warn', `None available provider found for ${providerType} (Model: ${requestedModel}) after checking fallback chain and model mapping.`); return null; } /** * Gets the fallback chain for a given provider type. * @param {string} providerType - The provider type to get fallback chain for. * @returns {Array} The fallback chain array, or empty array if not configured. */ getFallbackChain(providerType) { return this.fallbackChain[providerType] || []; } /** * Sets or updates the fallback chain for a provider type. * @param {string} providerType - The provider type to set fallback chain for. * @param {Array} fallbackTypes - Array of fallback provider types. */ setFallbackChain(providerType, fallbackTypes) { if (!Array.isArray(fallbackTypes)) { this._log('error', `Invalid fallbackTypes: must be an array`); return; } this.fallbackChain[providerType] = fallbackTypes; this._log('info', `Updated fallback chain for ${providerType}: ${fallbackTypes.join(' -> ')}`); } /** * Checks if all providers of a given type are unhealthy. * @param {string} providerType - The provider type to check. * @returns {boolean} True if all providers are unhealthy or disabled. */ isAllProvidersUnhealthy(providerType) { const providers = this.providerStatus[providerType] || []; if (providers.length === 0) { return true; } return providers.every(p => !p.config.isHealthy || p.config.isDisabled); } /** * Gets statistics about provider health for a given type. * @param {string} providerType - The provider type to get stats for. * @returns {Object} Statistics object with total, healthy, unhealthy, and disabled counts. */ getProviderStats(providerType) { const providers = this.providerStatus[providerType] || []; const stats = { total: providers.length, healthy: 0, unhealthy: 0, disabled: 0 }; for (const p of providers) { if (p.config.isDisabled) { stats.disabled++; } else if (p.config.isHealthy) { stats.healthy++; } else { stats.unhealthy++; } } return stats; } /** * Gets all available models across all provider pools, with optional format conversion. * @param {string} [endpointType] - Optional endpoint type for format conversion (OPENAI_MODEL_LIST or GEMINI_MODEL_LIST). * @returns {Promise} Formatted model list or raw array of model objects. */ async getAllAvailableModels(endpointType = null) { const allModels = []; // 获取所有已注册的提供商和号池中的提供商 const registeredProviders = getRegisteredProviders(); const allProviderTypes = Array.from(new Set([...registeredProviders])); for (const providerType of allProviderTypes) { if (this.providerStatus[providerType]) { let models = getProviderModels(providerType); // 如果硬编码的模型列表为空，或者该类型的提供商在号池中没有配置节点，尝试从服务获取 if (models.length === 0) { try { // 确定使用的配置：优先使用号池中第一个节点的配置，否则使用全局配置 let targetConfig = this.globalConfig; if (this.providerStatus[providerType] && this.providerStatus[providerType].length > 0) { targetConfig = this.providerStatus[providerType][0].config; } const tempConfig = { ...this.globalConfig, ...targetConfig, MODEL_PROVIDER: providerType }; const serviceAdapter = getServiceAdapter(tempConfig); if (typeof serviceAdapter.listModels === 'function') { const nativeModels = await serviceAdapter.listModels(); // 统一转换为 OpenAI 格式以便提取 ID const convertedData = convertData(nativeModels, 'modelList', providerType, MODEL_PROVIDER.OPENAI_CUSTOM); if (convertedData && Array.isArray(convertedData.data)) { const fetchedModels = convertedData.data.map(m => m.id); if (fetchedModels.length > 0) { models = fetchedModels; } } } } catch (err) { this._log('debug', `Failed to fetch model list for ${providerType} from service: ${err.message}`); // 保持原有的 models (可能是硬编码的空列表或 getProviderModels 返回的结果) } } for (const model of models) { allModels.push({ id: `${providerType}:${model}`, provider: providerType, model: model }); } } } // 如果没有指定 endpointType，返回原始数组 if (!endpointType) { return allModels; } // 根据 endpointType 转换为对应格式 if (endpointType === ENDPOINT_TYPE.OPENAI_MODEL_LIST) { // OpenAI 格式聚合 return { object: "list", data: allModels.map(m => ({ id: m.id, object: "model", created: Math.floor(Date.now() / 1000), owned_by: m.provider })) }; } else if (endpointType === ENDPOINT_TYPE.GEMINI_MODEL_LIST) { // Gemini 格式聚合 return { models: allModels.map(m => ({ name: `models/${m.id}`, baseModelId: m.model, version: "v1", displayName: `${m.model} (${m.provider})`, description: `Model ${m.model} provided by ${m.provider}`, supportedGenerationMethods: ["generateContent", "countTokens"] })) }; } // 默认返回空列表 return { data: [] }; } /** * 标记提供商需要刷新并推入刷新队列 * @param {string} providerType - 提供商类型 * @param {object} providerConfig - 提供商配置（包含 uuid） */ markProviderNeedRefresh(providerType, providerConfig) { if (!providerConfig?.uuid) { this._log('error', 'Invalid providerConfig in markProviderNeedRefresh'); return; } const provider = this._findProvider(providerType, providerConfig.uuid); if (provider) { // 防并发机制 A: 如果已经在刷新中，忽略请求 if (this.refreshingUuids.has(provider.uuid)) { this._log('debug', `Provider ${providerConfig.uuid} is already in refresh queue, ignoring duplicate request.`); return; } // 防并发机制 B: 如果 30 秒内刚刷新过，忽略请求（防止滞后的 401 错误导致重复刷新） const now = Date.now(); const lastRefreshTime = provider.config.lastRefreshTime || 0; if (now - lastRefreshTime < 30000) { this._log('info', `Provider ${providerConfig.uuid} was refreshed recently (${Math.round((now - lastRefreshTime)/1000)}s ago), ignoring refresh request.`); return; } provider.config.needsRefresh = true; this._log('info', `Marked provider ${providerConfig.uuid} as needsRefresh. Enqueuing...`); // 推入异步刷新队列 this._enqueueRefresh(providerType, provider, true); this._debouncedSave(providerType); } } /** * Marks a provider as unhealthy (e.g., after an API error). * @param {string} providerType - The type of the provider. * @param {object} providerConfig - The configuration of the provider to mark. * @param {string} [errorMessage] - Optional error message to store. */ markProviderUnhealthy(providerType, providerConfig, errorMessage = null) { if (!providerConfig?.uuid) { this._log('error', 'Invalid providerConfig in markProviderUnhealthy'); return; } const provider = this._findProvider(providerType, providerConfig.uuid); if (provider) { const wasHealthy = provider.config.isHealthy; const now = Date.now(); const lastErrorTime = provider.config.lastErrorTime ? new Date(provider.config.lastErrorTime).getTime() : 0; const errorWindowMs = 10000; // 10 秒窗口期 // 如果距离上次错误超过窗口期，重置错误计数 if (now - lastErrorTime > errorWindowMs) { provider.config.errorCount = 1; } else { provider.config.errorCount++; } provider.config.lastErrorTime = new Date().toISOString(); // 更新 lastUsed 时间，避免因 LRU 策略导致失败节点被重复选中 provider.config.lastUsed = new Date().toISOString(); // 只要报错，就清除刷新标记，由下次触发或健康检查决定是否需要刷新 provider.config.needsRefresh = false; provider.config.refreshCount = 0; // 保存错误信息 if (errorMessage) { provider.config.lastErrorMessage = errorMessage; } if (this.maxErrorCount > 0 && provider.config.errorCount >= this.maxErrorCount) { provider.config.isHealthy = false; // 健康状态变化日志 if (wasHealthy) { this._logHealthStatusChange(providerType, provider.config, 'healthy', 'unhealthy', errorMessage); } this._log('warn', `Marked provider as unhealthy: ${providerConfig.uuid} for type ${providerType}. Total errors: ${provider.config.errorCount}`); } this._debouncedSave(providerType); } } /** * Marks a provider as unhealthy immediately (without accumulating error count). * Used for definitive authentication errors like 401/403. * @param {string} providerType - The type of the provider. * @param {object} providerConfig - The configuration of the provider to mark. * @param {string} [errorMessage] - Optional error message to store. */ markProviderUnhealthyImmediately(providerType, providerConfig, errorMessage = null) { if (!providerConfig?.uuid) { this._log('error', 'Invalid providerConfig in markProviderUnhealthyImmediately'); return; } const provider = this._findProvider(providerType, providerConfig.uuid); if (provider) { const wasHealthy = provider.config.isHealthy; provider.config.isHealthy = false; provider.config.needsRefresh = false; // 报错时不健康，清除刷新标记，防止卡死 provider.config.refreshCount = 0; provider.config.errorCount = this.maxErrorCount; // Set to max to indicate definitive failure provider.config.lastErrorTime = new Date().toISOString(); provider.config.lastUsed = new Date().toISOString(); if (errorMessage) { provider.config.lastErrorMessage = errorMessage; } // 健康状态变化日志 if (wasHealthy) { this._logHealthStatusChange(providerType, provider.config, 'healthy', 'unhealthy', errorMessage); } this._log('warn', `Immediately marked provider as unhealthy: ${providerConfig.uuid} for type ${providerType}. Reason: ${errorMessage || 'Authentication error'}`); this._debouncedSave(providerType); } } /** * Marks a provider as unhealthy with a scheduled recovery time. * Used for quota exhaustion errors (402) where the quota will reset at a specific time. * @param {string} providerType - The type of the provider. * @param {object} providerConfig - The configuration of the provider to mark. * @param {string} [errorMessage] - Optional error message to store. * @param {Date|string} [recoveryTime] - Optional recovery time when the provider should be marked healthy again. */ markProviderUnhealthyWithRecoveryTime(providerType, providerConfig, errorMessage = null, recoveryTime = null) { if (!providerConfig?.uuid) { this._log('error', 'Invalid providerConfig in markProviderUnhealthyWithRecoveryTime'); return; } const provider = this._findProvider(providerType, providerConfig.uuid); if (provider) { provider.config.isHealthy = false; provider.config.needsRefresh = false; // 报错时不健康，清除刷新标记，防止卡死 provider.config.refreshCount = 0; provider.config.errorCount = this.maxErrorCount; // Set to max to indicate definitive failure provider.config.lastErrorTime = new Date().toISOString(); provider.config.lastUsed = new Date().toISOString(); if (errorMessage) { provider.config.lastErrorMessage = errorMessage; } // Set recovery time if provided if (recoveryTime) { const recoveryDate = recoveryTime instanceof Date ? recoveryTime : new Date(recoveryTime); provider.config.scheduledRecoveryTime = recoveryDate.toISOString(); this._log('warn', `Marked provider as unhealthy with recovery time: ${providerConfig.uuid} for type ${providerType}. Recovery at: ${recoveryDate.toISOString()}. Reason: ${errorMessage || 'Quota exhausted'}`); } else { this._log('warn', `Marked provider as unhealthy: ${providerConfig.uuid} for type ${providerType}. Reason: ${errorMessage || 'Quota exhausted'}`); } this._debouncedSave(providerType); } } /** * Marks a provider as healthy. * @param {string} providerType - The type of the provider. * @param {object} providerConfig - The configuration of the provider to mark. * @param {boolean} resetUsageCount - Whether to reset usage count (optional, default: false). * @param {string} [healthCheckModel] - Optional model name used for health check. */ markProviderHealthy(providerType, providerConfig, resetUsageCount = false, healthCheckModel = null) { if (!providerConfig?.uuid) { this._log('error', 'Invalid providerConfig in markProviderHealthy'); return; } const provider = this._findProvider(providerType, providerConfig.uuid); if (provider) { const wasHealthy = provider.config.isHealthy; provider.config.isHealthy = true; provider.config.errorCount = 0; provider.config.refreshCount = 0; provider.config.needsRefresh = false; provider.config.lastRefreshTime = Date.now(); // 标记为健康时也视为刚刷新完成 provider.config.lastErrorTime = null; provider.config.lastErrorMessage = null; provider.config._lastSelectionSeq = 0; // 更新健康检测信息 if (healthCheckModel) { provider.config.lastHealthCheckTime = new Date().toISOString(); provider.config.lastHealthCheckModel = healthCheckModel; } // 只有在明确要求重置使用计数时才重置 if (resetUsageCount) { provider.config.usageCount = 0; }else{ provider.config.usageCount++; provider.config.lastUsed = new Date().toISOString(); } // 健康状态变化日志 if (!wasHealthy) { this._logHealthStatusChange(providerType, provider.config, 'unhealthy', 'healthy', null); } this._log('info', `Marked provider as healthy: ${provider.config.uuid} for type ${providerType}${resetUsageCount ? ' (usage count reset)' : ''}`); this._debouncedSave(providerType); } } /** * 重置提供商的刷新状态（needsRefresh 和 refreshCount） * 并将其标记为健康，以便立即投入使用 * @param {string} providerType - 提供商类型 * @param {string} uuid - 提供商 UUID */ resetProviderRefreshStatus(providerType, uuid) { if (!providerType || !uuid) { this._log('error', 'Invalid parameters in resetProviderRefreshStatus'); return; } const provider = this._findProvider(providerType, uuid); if (provider) { provider.config.needsRefresh = false; provider.config.refreshCount = 0; provider.config.lastRefreshTime = Date.now(); // 显式重置时也更新刷新时间 // 更新为可用 provider.config.lastHealthCheckTime = new Date().toISOString(); // 标记为健康，以便立即投入使用 this._log('info', `Reset refresh status and marked healthy for provider ${uuid} (${providerType})`); this._debouncedSave(providerType); } } /** * 重置提供商的计数器（错误计数和使用计数） * @param {string} providerType - The type of the provider. * @param {object} providerConfig - The configuration of the provider to mark. */ resetProviderCounters(providerType, providerConfig) { if (!providerConfig?.uuid) { this._log('error', 'Invalid providerConfig in resetProviderCounters'); return; } const provider = this._findProvider(providerType, providerConfig.uuid); if (provider) { provider.config.errorCount = 0; provider.config.usageCount = 0; provider.config._lastSelectionSeq = 0; this._log('info', `Reset provider counters: ${provider.config.uuid} for type ${providerType}`); this._debouncedSave(providerType); } } /** * 禁用指定提供商 * @param {string} providerType - 提供商类型 * @param {object} providerConfig - 提供商配置 */ disableProvider(providerType, providerConfig) { if (!providerConfig?.uuid) { this._log('error', 'Invalid providerConfig in disableProvider'); return; } const provider = this._findProvider(providerType, providerConfig.uuid); if (provider) { provider.config.isDisabled = true; this._log('info', `Disabled provider: ${providerConfig.uuid} for type ${providerType}`); this._debouncedSave(providerType); } } /** * 启用指定提供商 * @param {string} providerType - 提供商类型 * @param {object} providerConfig - 提供商配置 */ enableProvider(providerType, providerConfig) { if (!providerConfig?.uuid) { this._log('error', 'Invalid providerConfig in enableProvider'); return; } const provider = this._findProvider(providerType, providerConfig.uuid); if (provider) { provider.config.isDisabled = false; this._log('info', `Enabled provider: ${providerConfig.uuid} for type ${providerType}`); this._debouncedSave(providerType); } } /** * 刷新指定提供商的 UUID * 用于在认证错误（如 401）时更换 UUID，以便重新尝试 * @param {string} providerType - 提供商类型 * @param {object} providerConfig - 提供商配置（包含当前 uuid） * @returns {string|null} 新的 UUID，如果失败则返回 null */ refreshProviderUuid(providerType, providerConfig) { if (!providerConfig?.uuid) { this._log('error', 'Invalid providerConfig in refreshProviderUuid'); return null; } const provider = this._findProvider(providerType, providerConfig.uuid); if (provider) { const oldUuid = provider.config.uuid; // 生成新的 UUID const newUuid = 'xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx'.replace(/[xy]/g, function(c) { const r = Math.random() * 16 | 0; const v = c === 'x' ? r : (r & 0x3 | 0x8); return v.toString(16); }); // 更新 provider 的 UUID provider.uuid = newUuid; provider.config.uuid = newUuid; // 同时更新 providerPools 中的原始数据 const poolArray = this.providerPools[providerType]; if (poolArray) { const originalProvider = poolArray.find(p => p.uuid === oldUuid); if (originalProvider) { originalProvider.uuid = newUuid; } } this._log('info', `Refreshed provider UUID: ${oldUuid} -> ${newUuid} for type ${providerType}`); this._debouncedSave(providerType); return newUuid; } this._log('warn', `Provider not found for UUID refresh: ${providerConfig.uuid} in ${providerType}`); return null; } /** * 检查并恢复已到恢复时间的提供商 * @param {string} [providerType] - 可选，指定要检查的提供商类型。如果不提供，检查所有类型 * @private */ _checkAndRecoverScheduledProviders(providerType = null) { const now = new Date(); const typesToCheck = providerType ? [providerType] : Object.keys(this.providerStatus); for (const type of typesToCheck) { const providers = this.providerStatus[type] || []; for (const providerStatus of providers) { const config = providerStatus.config; // 检查是否有 scheduledRecoveryTime 且已到恢复时间 if (config.scheduledRecoveryTime && !config.isHealthy) { const recoveryTime = new Date(config.scheduledRecoveryTime); if (now >= recoveryTime) { this._log('info', `Auto-recovering provider ${config.uuid} (${type}). Scheduled recovery time reached: ${recoveryTime.toISOString()}`); // 恢复健康状态 config.isHealthy = true; config.errorCount = 0; config.lastErrorTime = null; config.lastErrorMessage = null; config.scheduledRecoveryTime = null; // 清除恢复时间 // 保存更改 this._debouncedSave(type); } } } } } /** * Performs health checks on selected providers. * Respects SCHEDULED_HEALTH_CHECK.providerTypes configuration. * * 设计决策：如果没有选择任何 provider types，则不进行检查任何 provider。 * 这是有意为之的设计 - 如果用户没有明确选择，则不需要自动健康检查。 * 区别于原来的逻辑（检查所有 provider），现在的行为更符合用户预期。 */ async performHealthChecks() { const scheduledConfig = this.globalConfig?.SCHEDULED_HEALTH_CHECK; const selectedProviderTypes = scheduledConfig?.providerTypes; // 如果没有选择任何 provider types，不进行检查 // 设计决策：如果用户没有选择任何 provider，明确不执行健康检查是合理的 if (!Array.isArray(selectedProviderTypes) || selectedProviderTypes.length === 0) { return; } this._log('info', 'Performing health checks on selected providers...'); const now = new Date(); // 首先检查并恢复已到恢复时间的提供商 this._checkAndRecoverScheduledProviders(); for (const providerType in this.providerStatus) { // Only check selected provider types if (!selectedProviderTypes.includes(providerType)) { continue; } for (const providerStatus of this.providerStatus[providerType]) { const providerConfig = providerStatus.config; // 如果提供商有 scheduledRecoveryTime 且未到恢复时间，跳过健康检查 if (providerConfig.scheduledRecoveryTime && !providerConfig.isHealthy) { const recoveryTime = new Date(providerConfig.scheduledRecoveryTime); if (now < recoveryTime) { this._log('debug', `Skipping health check for ${providerConfig.uuid} (${providerType}). Waiting for scheduled recovery at ${recoveryTime.toISOString()}`); continue; } } // Only attempt to health check unhealthy providers after a certain interval if (!providerStatus.config.isHealthy && providerStatus.config.lastErrorTime && (now.getTime() - new Date(providerStatus.config.lastErrorTime).getTime() < this.healthCheckInterval)) { this._log('debug', `Skipping health check for ${providerConfig.uuid} (${providerType}). Last error too recent.`); continue; } try { // Perform actual health check based on provider type const healthResult = await this._checkProviderHealth(providerType, providerConfig); if (healthResult === null) { this._log('debug', `Health check for ${providerConfig.uuid} (${providerType}) skipped: Check not implemented.`); this.resetProviderCounters(providerType, providerConfig); continue; } if (healthResult.success) { if (!providerStatus.config.isHealthy) { // Provider was unhealthy but is now healthy // 恢复健康时不重置使用计数，保持原有值 this.markProviderHealthy(providerType, providerConfig, true, healthResult.modelName); this._log('info', `Health check for ${providerConfig.uuid} (${providerType}): Marked Healthy (actual check)`); } else { // Provider was already healthy and still is // 只在初始化时重置使用计数 this.markProviderHealthy(providerType, providerConfig, true, healthResult.modelName); this._log('debug', `Health check for ${providerConfig.uuid} (${providerType}): Still Healthy`); } } else { // Provider is not healthy this._log('warn', `Health check for ${providerConfig.uuid} (${providerType}) failed: ${healthResult.errorMessage || 'Provider is not responding correctly.'}`); this.markProviderUnhealthy(providerType, providerConfig, healthResult.errorMessage); // 更新健康检测时间和模型（即使失败也记录） providerStatus.config.lastHealthCheckTime = new Date().toISOString(); if (healthResult.modelName) { providerStatus.config.lastHealthCheckModel = healthResult.modelName; } } } catch (error) { this._log('error', `Health check for ${providerConfig.uuid} (${providerType}) failed: ${error.message}`); // If a health check fails, mark it unhealthy, which will update error count and lastErrorTime this.markProviderUnhealthy(providerType, providerConfig, error.message); } } } } /** * Performs scheduled health checks on all providers. * This method is designed to be called periodically to proactively check provider health. * It respects provider-level isDisabled flag. */ async performScheduledHealthChecks() { const scheduledConfig = this.globalConfig?.SCHEDULED_HEALTH_CHECK; const checkStartTime = Date.now(); // Check if scheduled health checks are disabled if (!scheduledConfig?.enabled) { this._log('debug', '[ScheduledHealthCheck] Scheduled health checks are disabled via configuration'); return; } // Get selected provider types let selectedProviderTypes = scheduledConfig?.providerTypes; // Validate providerTypes is an array if (!Array.isArray(selectedProviderTypes) || selectedProviderTypes.length === 0) { this._log('info', '[ScheduledHealthCheck] No provider types selected, skipping health check'); return; } // Count providers to be checked let totalProviders = 0; let providersToCheck = []; for (const providerType in this.providerStatus) { // Only check selected provider types if (!selectedProviderTypes.includes(providerType)) { this._log('debug', `[ScheduledHealthCheck] Skipping provider type ${providerType}: not in selected types`); continue; } for (const provider of this.providerStatus[providerType]) { // Skip manually disabled providers if (provider.config.isDisabled === true) { this._log('debug', `[ScheduledHealthCheck] Skipping ${provider.config.uuid} (${providerType}): manually disabled`); continue; } totalProviders++; providersToCheck.push({ providerType, provider, uuid: provider.config.uuid, customName: provider.config.customName }); } } this._log('info', `[ScheduledHealthCheck] Starting scheduled health checks: ${totalProviders} provider(s) to check (interval: ${scheduledConfig.interval}ms, types: ${selectedProviderTypes.join(', ')})`); let successCount = 0; let failCount = 0; for (const { providerType, provider, uuid, customName } of providersToCheck) { // Skip if provider became disabled during iteration if (provider.config.isDisabled === true) { continue; } const providerCheckStart = Date.now(); const checkModelName = provider.config.checkModelName || ProviderPoolManager.DEFAULT_HEALTH_CHECK_MODELS[providerType] || 'unknown'; const displayName = customName || uuid.substring(0, 8); try { // Perform health check (health check is based on providerTypes configuration, not per-provider checkHealth flag) const result = await this._checkProviderHealth(providerType, provider.config); const checkDuration = Date.now() - providerCheckStart; if (!result.success) { // Provider is unhealthy failCount++; this._log('warn', `[ScheduledHealthCheck] ${displayName} (${providerType}) FAILED: ${result.errorMessage || 'Provider is not responding correctly.'} (${checkDuration}ms)`); this.markProviderUnhealthyImmediately(providerType, provider.config, result.errorMessage); } else { // Provider is healthy successCount++; this._log('info', `[ScheduledHealthCheck] ${displayName} (${providerType}) PASSED: model=${result.modelName || checkModelName} (${checkDuration}ms)`); this.markProviderHealthy(providerType, provider.config, false, result.modelName); } } catch (error) { const checkDuration = Date.now() - providerCheckStart; failCount++; this._log('error', `[ScheduledHealthCheck] ${displayName} (${providerType}) EXCEPTION: ${error.message} (${checkDuration}ms)`); this.markProviderUnhealthyImmediately(providerType, provider.config, error.message); } } const totalDuration = Date.now() - checkStartTime; this._log('info', `[ScheduledHealthCheck] Completed: ${successCount} passed, ${failCount} failed, ${totalDuration}ms total`); } /** * 构建健康检查请求（返回多种格式用于重试） * @private * @returns {Array} 请求格式数组，按优先级排序 */ _buildHealthCheckRequests(providerType, modelName) { const baseMessage = { role: 'user', content: 'Hi' }; const requests = []; // Gemini 使用 contents 格式 if (providerType.startsWith('gemini')) { requests.push({ contents: [{ role: 'user', parts: [{ text: baseMessage.content }] }] }); return requests; } // Kiro OAuth 只支持 messages 格式 if (providerType.startsWith('claude-kiro')) { requests.push({ messages: [baseMessage], model: modelName, max_tokens: 1 }); return requests; } // OpenAI Custom Responses 使用特殊格式 if (providerType === MODEL_PROVIDER.OPENAI_CUSTOM_RESPONSES) { requests.push({ input: [baseMessage], model: modelName }); return requests; } // 其他提供商（OpenAI、Claude、Qwen）使用标准 messages 格式 requests.push({ messages: [baseMessage], model: modelName }); return requests; } /** * Performs an actual health check for a specific provider. * * 设计决策：不检查 providerConfig.checkHealth 标志。 * 健康检查是否执行由上层调用方（performScheduledHealthChecks / performHealthChecks） * 通过 providerTypes 数组来决定，不在每个 provider 级别控制。 * 这样简化了逻辑，避免 per-provider 的 checkHealth flag 变得无用。 * * @param {string} providerType - The type of the provider. * @param {object} providerConfig - The configuration of the provider to check. * @returns {Promise<{success: boolean, modelName: string, errorMessage: string}>} - Health check result object. */ async _checkProviderHealth(providerType, providerConfig) { // 确定健康检查使用的模型名称 const modelName = providerConfig.checkModelName || ProviderPoolManager.DEFAULT_HEALTH_CHECK_MODELS[providerType]; if (!modelName) { this._log('warn', `Unknown provider type for health check: ${providerType}. Please check DEFAULT_HEALTH_CHECK_MODELS.`); return { success: false, modelName: null, errorMessage: `Unknown provider type '${providerType}'. No default health check model configured.` }; } // ========== 实际 API 健康检查（带超时保护）========== const tempConfig = { ...providerConfig, MODEL_PROVIDER: providerType }; const serviceAdapter = getServiceAdapter(tempConfig); // 获取所有可能的请求格式 const healthCheckRequests = this._buildHealthCheckRequests(providerType, modelName); // 健康检查超时时间（15秒，避免长时间阻塞） const healthCheckTimeout = 15000; let lastError = null; // 重试机制：尝试不同的请求格式 for (let i = 0; i < healthCheckRequests.length; i++) { const healthCheckRequest = healthCheckRequests[i]; const abortController = new AbortController(); const timeoutId = setTimeout(() => abortController.abort(), healthCheckTimeout); try { // 尝试将 signal 注入请求体，供支持的适配器使用 const requestWithSignal = { ...healthCheckRequest, // signal: abortController.signal }; await serviceAdapter.generateContent(modelName, requestWithSignal); clearTimeout(timeoutId); // 注意：使用量计数由调用方处理（performScheduledHealthChecks/performHealthChecks） // 这里只返回成功结果，让调用方统一处理状态更新和计数 return { success: true, modelName, errorMessage: null }; } catch (error) { clearTimeout(timeoutId); lastError = error; } } // 所有尝试都失败 this._log('warn', `[HealthCheck] ${providerType} failed after ${healthCheckRequests.length} attempts: ${lastError?.message}`); return { success: false, modelName, errorMessage: lastError?.message || 'All health check attempts failed' }; } /** * 优化1: 添加防抖保存方法 * 延迟保存操作，避免频繁的文件 I/O * @private */ _debouncedSave(providerType) { // 将待保存的 providerType 添加到集合中 this.pendingSaves.add(providerType); // 清除之前的定时器 if (this.saveTimer) { clearTimeout(this.saveTimer); } // 设置新的定时器 this.saveTimer = setTimeout(() => { this._flushPendingSaves(); }, this.saveDebounceTime); } /** * 批量保存所有待保存的 providerType（优化为单次文件写入） * @private */ async _flushPendingSaves() { const typesToSave = Array.from(this.pendingSaves); if (typesToSave.length === 0) return; this.pendingSaves.clear(); this.saveTimer = null; try { const filePath = this.globalConfig.PROVIDER_POOLS_FILE_PATH || 'configs/provider_pools.json'; let currentPools = {}; // 一次性读取文件 try { const fileContent = await fs.promises.readFile(filePath, 'utf8'); currentPools = JSON.parse(fileContent); } catch (readError) { if (readError.code === 'ENOENT') { this._log('info', 'configs/provider_pools.json does not exist, creating new file.'); } else { throw readError; } } // 更新所有待保存的 providerType for (const providerType of typesToSave) { if (this.providerStatus[providerType]) { currentPools[providerType] = this.providerStatus[providerType].map(p => { // Convert Date objects to ISOString if they exist const config = { ...p.config }; if (config.lastUsed instanceof Date) { config.lastUsed = config.lastUsed.toISOString(); } if (config.lastErrorTime instanceof Date) { config.lastErrorTime = config.lastErrorTime.toISOString(); } if (config.lastHealthCheckTime instanceof Date) { config.lastHealthCheckTime = config.lastHealthCheckTime.toISOString(); } return config; }); } else { this._log('warn', `Attempted to save unknown providerType: ${providerType}`); } } // 一次性写入文件 await fs.promises.writeFile(filePath, JSON.stringify(currentPools, null, 2), 'utf8'); this._log('info', `configs/provider_pools.json updated successfully for types: ${typesToSave.join(', ')}`); } catch (error) { this._log('error', `Failed to write provider_pools.json: ${error.message}`); } } }