import express from 'express'; import { fal } from '@fal-ai/client'; // --- Key Management --- const FAL_KEY_STRING = process.env.FAL_KEY; const API_KEY = process.env.API_KEY; // 自定义 API Key 环境变量保持不变 if (!FAL_KEY_STRING) { console.error("Error: FAL_KEY environment variable is not set."); process.exit(1); } // 解析 FAL_KEY 字符串为数组,去除空白并过滤空值 const falKeys = FAL_KEY_STRING.split(',') .map(key => key.trim()) .filter(key => key.length > 0); if (falKeys.length === 0) { console.error("Error: FAL_KEY environment variable is set, but no valid keys were found after parsing."); process.exit(1); } console.log(`Loaded ${falKeys.length} Fal AI Keys initially.`); // 不再需要 currentFalKeyIndex // let currentFalKeyIndex = 0; // --- End Key Management --- if (!API_KEY) { console.error("Error: API_KEY environment variable is not set."); process.exit(1); } const app = express(); app.use(express.json({ limit: '50mb' })); app.use(express.urlencoded({ extended: true, limit: '50mb' })); const PORT = process.env.PORT || 3000; // API Key 鉴权中间件 (保持不变) const apiKeyAuth = (req, res, next) => { const authHeader = req.headers['authorization']; if (!authHeader) { console.warn('Unauthorized: No Authorization header provided'); return res.status(401).json({ error: 'Unauthorized: No API Key provided' }); } const authParts = authHeader.split(' '); if (authParts.length !== 2 || authParts[0].toLowerCase() !== 'bearer') { console.warn('Unauthorized: Invalid Authorization header format'); return res.status(401).json({ error: 'Unauthorized: Invalid Authorization header format' }); } const providedKey = authParts[1]; if (providedKey !== API_KEY) { console.warn('Unauthorized: Invalid API Key'); return res.status(401).json({ error: 'Unauthorized: Invalid API Key' }); } next(); }; // 应用 API Key 鉴权中间件到所有 API 路由 (保持不变) app.use(['/v1/models', '/v1/chat/completions'], apiKeyAuth); // === 全局定义限制 === (保持不变) const PROMPT_LIMIT = 4800; const SYSTEM_PROMPT_LIMIT = 4800; // === 限制定义结束 === // 定义 fal-ai/any-llm 支持的模型列表 (保持不变) const FAL_SUPPORTED_MODELS = [ "anthropic/claude-3.7-sonnet", "anthropic/claude-3.5-sonnet", "anthropic/claude-3-5-haiku", "anthropic/claude-3-haiku", "google/gemini-pro-1.5", "google/gemini-flash-1.5", "google/gemini-flash-1.5-8b", "google/gemini-2.0-flash-001", "meta-llama/llama-3.2-1b-instruct", "meta-llama/llama-3.2-3b-instruct", "meta-llama/llama-3.1-8b-instruct", "meta-llama/llama-3.1-70b-instruct", "openai/gpt-4o-mini", "openai/gpt-4o", "deepseek/deepseek-r1", "meta-llama/llama-4-maverick", "meta-llama/llama-4-scout" ]; // Helper function to get owner from model ID (保持不变) const getOwner = (modelId) => { if (modelId && modelId.includes('/')) { return modelId.split('/')[0]; } return 'fal-ai'; } // GET /v1/models endpoint (保持不变) app.get('/v1/models', (req, res) => { console.log("Received request for GET /v1/models"); try { const modelsData = FAL_SUPPORTED_MODELS.map(modelId => ({ id: modelId, object: "model", created: 1700000000, owned_by: getOwner(modelId) })); res.json({ object: "list", data: modelsData }); console.log("Successfully returned model list."); } catch (error) { console.error("Error processing GET /v1/models:", error); res.status(500).json({ error: "Failed to retrieve model list." }); } }); // === convertMessagesToFalPrompt 函数 (保持不变) === function convertMessagesToFalPrompt(messages) { let fixed_system_prompt_content = ""; const conversation_message_blocks = []; // console.log(`Original messages count: ${messages.length}`); // Reduced logging verbosity // 1. 分离 System 消息,格式化 User/Assistant 消息 for (const message of messages) { let content = (message.content === null || message.content === undefined) ? "" : String(message.content); switch (message.role) { case 'system': fixed_system_prompt_content += `System: ${content}\n\n`; break; case 'user': conversation_message_blocks.push(`Human: ${content}\n\n`); break; case 'assistant': conversation_message_blocks.push(`Assistant: ${content}\n\n`); break; default: console.warn(`Unsupported role: ${message.role}`); continue; } } // 2. 截断合并后的 system 消息(如果超长) if (fixed_system_prompt_content.length > SYSTEM_PROMPT_LIMIT) { const originalLength = fixed_system_prompt_content.length; fixed_system_prompt_content = fixed_system_prompt_content.substring(0, SYSTEM_PROMPT_LIMIT); console.warn(`Combined system messages truncated from ${originalLength} to ${SYSTEM_PROMPT_LIMIT}`); } fixed_system_prompt_content = fixed_system_prompt_content.trim(); // 3. 计算 system_prompt 中留给对话历史的剩余空间 let space_occupied_by_fixed_system = 0; if (fixed_system_prompt_content.length > 0) { space_occupied_by_fixed_system = fixed_system_prompt_content.length + 4; // 预留 \n\n...\n\n 的长度 } const remaining_system_limit = Math.max(0, SYSTEM_PROMPT_LIMIT - space_occupied_by_fixed_system); // console.log(`Trimmed fixed system prompt length: ${fixed_system_prompt_content.length}. Approx remaining system history limit: ${remaining_system_limit}`); // 4. 反向填充 User/Assistant 对话历史 const prompt_history_blocks = []; const system_prompt_history_blocks = []; let current_prompt_length = 0; let current_system_history_length = 0; let promptFull = false; let systemHistoryFull = (remaining_system_limit <= 0); // console.log(`Processing ${conversation_message_blocks.length} user/assistant messages for recency filling.`); for (let i = conversation_message_blocks.length - 1; i >= 0; i--) { const message_block = conversation_message_blocks[i]; const block_length = message_block.length; if (promptFull && systemHistoryFull) { // console.log(`Both prompt and system history slots full. Omitting older messages from index ${i}.`); break; } // 优先尝试放入 prompt if (!promptFull) { if (current_prompt_length + block_length <= PROMPT_LIMIT) { prompt_history_blocks.unshift(message_block); current_prompt_length += block_length; continue; } else { promptFull = true; // console.log(`Prompt limit (${PROMPT_LIMIT}) reached. Trying system history slot.`); } } // 如果 prompt 满了,尝试放入 system_prompt 的剩余空间 if (!systemHistoryFull) { if (current_system_history_length + block_length <= remaining_system_limit) { system_prompt_history_blocks.unshift(message_block); current_system_history_length += block_length; continue; } else { systemHistoryFull = true; // console.log(`System history limit (${remaining_system_limit}) reached.`); } } } // 5. *** 组合最终的 prompt 和 system_prompt (包含分隔符逻辑) *** const system_prompt_history_content = system_prompt_history_blocks.join('').trim(); const final_prompt = prompt_history_blocks.join('').trim(); const SEPARATOR = "\n\n-------下面是比较早之前的对话内容-----\n\n"; let final_system_prompt = ""; const hasFixedSystem = fixed_system_prompt_content.length > 0; const hasSystemHistory = system_prompt_history_content.length > 0; if (hasFixedSystem && hasSystemHistory) { final_system_prompt = fixed_system_prompt_content + SEPARATOR + system_prompt_history_content; // console.log("Combining fixed system prompt and history with separator."); } else if (hasFixedSystem) { final_system_prompt = fixed_system_prompt_content; // console.log("Using only fixed system prompt."); } else if (hasSystemHistory) { final_system_prompt = system_prompt_history_content; // console.log("Using only history in system prompt slot."); } const result = { system_prompt: final_system_prompt, prompt: final_prompt }; // console.log(`Final system_prompt length (Sys+Separator+Hist): ${result.system_prompt.length}`); // console.log(`Final prompt length (Hist): ${result.prompt.length}`); return result; } // === convertMessagesToFalPrompt 函数结束 === // POST /v1/chat/completions endpoint (随机 Key + 失败排除) app.post('/v1/chat/completions', async (req, res) => { const { model, messages, stream = false, reasoning = false, ...restOpenAIParams } = req.body; const requestId = `req-${Date.now()}`; console.log(`[${requestId}] Received chat completion request for model: ${model}, stream: ${stream}. Strategy: Random key with exclusion.`); if (!FAL_SUPPORTED_MODELS.includes(model)) { console.warn(`[${requestId}] Warning: Requested model '${model}' is not in the explicitly supported list.`); } if (!model || !messages || !Array.isArray(messages) || messages.length === 0) { console.error(`[${requestId}] Invalid request parameters:`, { model, messages: Array.isArray(messages) ? messages.length : typeof messages }); return res.status(400).json({ error: 'Missing or invalid parameters: model and messages array are required.' }); } let lastError = null; let success = false; let attempt = 0; const maxAttempts = falKeys.length; // Safety limit // *** 為此請求創建可用的 Key 列表副本 *** let availableKeysForRequest = [...falKeys]; // 准备 Fal Input (只需要准备一次) const { prompt, system_prompt } = convertMessagesToFalPrompt(messages); const falInput = { model: model, prompt: prompt, ...(system_prompt && { system_prompt: system_prompt }), reasoning: !!reasoning, }; console.log(`[${requestId}] Fal Input prepared. System Prompt Length: ${system_prompt?.length || 0}, Prompt Length: ${prompt?.length || 0}`); // *** 重试循环:只要还有可用的 Key 且未达最大尝试次数 *** while (availableKeysForRequest.length > 0 && attempt < maxAttempts) { attempt++; // *** 隨機選擇一個 Key *** const randomIndex = Math.floor(Math.random() * availableKeysForRequest.length); const selectedFalKey = availableKeysForRequest[randomIndex]; // Mask key in logs for security const maskedKey = selectedFalKey.length > 8 ? `${selectedFalKey.substring(0, 4)}...${selectedFalKey.substring(selectedFalKey.length - 4)}` : selectedFalKey; console.log(`[${requestId}] Attempt ${attempt}/${maxAttempts}: Trying random key (masked: ${maskedKey}). ${availableKeysForRequest.length} keys available.`); try { fal.config({ credentials: selectedFalKey }); if (stream) { // --- 流式处理 --- res.setHeader('Content-Type', 'text/event-stream; charset=utf-8'); res.setHeader('Cache-Control', 'no-cache'); res.setHeader('Connection', 'keep-alive'); res.setHeader('Access-Control-Allow-Origin', '*'); let previousOutput = ''; let firstEventProcessed = false; let streamFailedMidway = false; let keyConfirmedWorking = false; const falStream = await fal.stream("fal-ai/any-llm", { input: falInput }); try { for await (const event of falStream) { const currentOutput = (event && typeof event.output === 'string') ? event.output : ''; const isPartial = (event && typeof event.partial === 'boolean') ? event.partial : true; const errorInfo = (event && event.error) ? event.error : null; const eventStatus = errorInfo?.status; if (errorInfo) { console.warn(`[${requestId}] Error in stream event (Key: ${maskedKey}):`, errorInfo); lastError = errorInfo; if (!firstEventProcessed && (eventStatus === 401 || eventStatus === 403 || eventStatus === 429)) { console.warn(`[${requestId}] Key ${maskedKey} failed (${eventStatus}) on first event. Excluding it for this request.`); availableKeysForRequest.splice(randomIndex, 1); // 从可用列表移除 console.log(`[${requestId}] ${availableKeysForRequest.length} keys remaining for this request.`); break; // 中断内部循环,外部循环将尝试下一个随机 key } else { console.error(`[${requestId}] Unrecoverable stream error or error after stream start (Key: ${maskedKey}).`); streamFailedMidway = true; if (!res.headersSent) { res.status(500).json({ object: "error", message: `Fal Stream Error: ${JSON.stringify(errorInfo)}`, type:"fal_stream_error"}); } else if (!res.writableEnded) { const errorChunk = { id: `chatcmpl-${Date.now()}-error`, object: "chat.completion.chunk", created: Math.floor(Date.now() / 1000), model: model, choices: [{ index: 0, delta: {}, finish_reason: "error", message: { role: 'assistant', content: `Fal Stream Error: ${JSON.stringify(errorInfo)}` } }] }; res.write(`data: ${JSON.stringify(errorChunk)}\n\n`); } break; // 中断内部循环 } } if (!keyConfirmedWorking && !errorInfo) { success = true; keyConfirmedWorking = true; console.log(`[${requestId}] Key ${maskedKey} confirmed working (stream).`); if (!res.headersSent) { res.flushHeaders(); console.log(`[${requestId}] Stream headers flushed.`); } firstEventProcessed = true; } if (!errorInfo) { let deltaContent = ''; if (currentOutput.startsWith(previousOutput)) { deltaContent = currentOutput.substring(previousOutput.length); } else if (currentOutput.length > 0) { deltaContent = currentOutput; previousOutput = ''; } previousOutput = currentOutput; if (deltaContent || !isPartial) { const openAIChunk = { id: `chatcmpl-${Date.now()}`, object: "chat.completion.chunk", created: Math.floor(Date.now() / 1000), model: model, choices: [{ index: 0, delta: { content: deltaContent }, finish_reason: isPartial === false ? "stop" : null }] }; if (!res.writableEnded) { res.write(`data: ${JSON.stringify(openAIChunk)}\n\n`); } } } } // End `for await...of` loop if (streamFailedMidway) { if (!res.writableEnded) { res.write(`data: [DONE]\n\n`); res.end(); } break; // 中断外部循环,因为流在中途失败 } else if (keyConfirmedWorking) { if (!res.writableEnded) { res.write(`data: [DONE]\n\n`); res.end(); } break; // 中断外部循环,因为成功 } // 如果内部循环因为首个事件是 key 错误而中断,外部循环会继续 } catch (streamProcessingError) { console.error(`[${requestId}] Error during fal stream processing loop logic (Key: ${maskedKey}):`, streamProcessingError); lastError = streamProcessingError; if (!res.headersSent) { res.status(500).json({ object: "error", message: `Proxy Stream Processing Error: ${streamProcessingError.message}`, type:"proxy_internal_error"}); } else if (!res.writableEnded) { try { res.write(`data: ${JSON.stringify({ error: { message: "Proxy Stream processing error", type: "proxy_internal_error", details: streamProcessingError.message } })}\n\n`); res.write(`data: [DONE]\n\n`); res.end(); } catch (finalError) { if (!res.writableEnded) { res.end(); } } } break; // 中断外部循环 } // 如果流成功或中途失败,外部循环会 break;如果因首个 key 错误而中断内部循环,则外部循环继续 if (success || streamFailedMidway) { break; } } else { // --- 非流式处理 --- console.log(`[${requestId}] Executing non-stream request (Key: ${maskedKey})...`); const result = await fal.subscribe("fal-ai/any-llm", { input: falInput, logs: true }); if (result && result.error) { console.error(`[${requestId}] Fal-ai returned a business error (Key: ${maskedKey}):`, result.error); lastError = new Error(`Fal-ai error: ${JSON.stringify(result.error)}`); lastError.status = result.status || 500; lastError.type = "fal_ai_error"; break; // 业务错误,中断重试 } console.log(`[${requestId}] Received non-stream result (Key: ${maskedKey}).`); success = true; // 标记成功 const openAIResponse = { id: `chatcmpl-${result.requestId || Date.now()}`, object: "chat.completion", created: Math.floor(Date.now() / 1000), model: model, choices: [{ index: 0, message: { role: "assistant", content: result.output || "" }, finish_reason: "stop" }], usage: { prompt_tokens: null, completion_tokens: null, total_tokens: null }, system_fingerprint: null, ...(result.reasoning && { fal_reasoning: result.reasoning }), }; res.json(openAIResponse); break; // 成功,中断外部循环 } } catch (error) { // Catch errors from fal.config, fal.stream/subscribe setup lastError = error; const status = error?.status; const errorMessage = error?.body?.detail || error?.message || 'Unknown setup error'; console.warn(`[${requestId}] Attempt ${attempt} with key ${maskedKey} failed during setup. Status: ${status || 'N/A'}, Message: ${errorMessage}`); // console.error("Setup Error details:", error); // Log full error for debug if needed if (status === 401 || status === 403 || status === 429) { console.warn(`[${requestId}] Key ${maskedKey} failed (${status}) during setup. Excluding it for this request.`); availableKeysForRequest.splice(randomIndex, 1); // 从可用列表移除 console.log(`[${requestId}] ${availableKeysForRequest.length} keys remaining for this request.`); // 继续外部循环 } else { console.error(`[${requestId}] Unrecoverable setup error encountered (Key: ${maskedKey}). Status: ${status || 'N/A'}. Stopping retries.`); break; // 中断外部循环 } } } // --- 结束重试循环 --- // 如果循环结束了还没有成功 if (!success) { console.error(`[${requestId}] All attempts failed or an unrecoverable error occurred. No available keys left or max attempts reached.`); if (!res.headersSent) { const statusCode = lastError?.status || 503; const errorMessage = (lastError instanceof Error) ? lastError.message : JSON.stringify(lastError); const detailMessage = lastError?.body?.detail || errorMessage; const errorType = lastError?.type || (statusCode === 401 || statusCode === 403 || statusCode === 429 ? "key_error" : "proxy_error"); console.error(`[${requestId}] Sending final error response. Status: ${statusCode}, Type: ${errorType}, Message: ${detailMessage}`); res.status(statusCode).json({ object: "error", message: `All attempts failed. Last error: ${detailMessage}`, type: errorType, param: null, code: statusCode === 429 ? "rate_limit_exceeded" : (statusCode === 401 || statusCode === 403 ? "invalid_api_key" : "service_unavailable") }); } else if (!res.writableEnded) { console.error(`[${requestId}] Headers potentially sent, but request failed. Ending stream.`); try { res.write(`data: [DONE]\n\n`); res.end(); } catch (e) { if (!res.writableEnded) res.end(); } } else { console.error(`[${requestId}] Request failed, but response stream was already fully ended.`); } } }); // 启动服务器 app.listen(PORT, () => { console.log(`===================================================`); console.log(` Fal OpenAI Proxy Server (Random Key + Exclusion Strategy)`); // 更新描述 console.log(` Listening on port: ${PORT}`); console.log(` Loaded ${falKeys.length} Fal AI Keys initially.`); console.log(` Using Limits: System Prompt=${SYSTEM_PROMPT_LIMIT}, Prompt=${PROMPT_LIMIT}`); console.log(` API Key Auth Enabled: ${API_KEY ? 'Yes' : 'No'}`); console.log(` Chat Completions Endpoint: POST http://localhost:${PORT}/v1/chat/completions`); console.log(` Models Endpoint: GET http://localhost:${PORT}/v1/models`); console.log(`===================================================`); }); // 根路径响应 app.get('/', (req, res) => { res.send('Fal OpenAI Proxy (Random Key + Exclusion Strategy) is running.'); // 更新描述 });