Spaces:
Sleeping
Sleeping
Update server.js
Browse files
server.js
CHANGED
|
@@ -1,48 +1,44 @@
|
|
| 1 |
import express from 'express';
|
| 2 |
import { fal } from '@fal-ai/client';
|
| 3 |
|
| 4 |
-
// ---
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
// === 全局定义限制 ===
|
| 15 |
-
const PROMPT_LIMIT = 4800;
|
| 16 |
-
const SYSTEM_PROMPT_LIMIT = 4800;
|
| 17 |
-
// === 限制定义结束 ===
|
| 18 |
-
|
| 19 |
-
// API Key 鉴权中间件
|
| 20 |
-
const apiKeyAuth = (req, res, next) => {
|
| 21 |
-
const authHeader = req.headers['authorization'];
|
| 22 |
-
|
| 23 |
-
if (!authHeader) {
|
| 24 |
-
console.warn('未提供 Authorization 头');
|
| 25 |
-
return res.status(401).json({ error: '未提供 API Key' });
|
| 26 |
-
}
|
| 27 |
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
}
|
| 33 |
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
next();
|
| 43 |
-
};
|
| 44 |
|
| 45 |
-
|
|
|
|
|
|
|
| 46 |
|
| 47 |
// NOTE: We will configure fal client per request now, so initial global config is removed.
|
| 48 |
|
|
@@ -84,6 +80,23 @@ function getNextKey() {
|
|
| 84 |
return null; // All keys are currently failed and within cooldown
|
| 85 |
}
|
| 86 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 87 |
/**
|
| 88 |
* Marks a specific key as failed.
|
| 89 |
* @param {object} keyInfo - The key info object to mark as failed.
|
|
@@ -97,25 +110,13 @@ function markKeyFailed(keyInfo) {
|
|
| 97 |
}
|
| 98 |
}
|
| 99 |
|
|
|
|
| 100 |
/**
|
| 101 |
* Determines if an error likely indicates an API key issue (auth, quota, etc.).
|
| 102 |
* This needs refinement based on actual errors from fal.ai.
|
| 103 |
* @param {Error} error - The error object caught from the fal client.
|
| 104 |
* @returns {boolean} - True if the error suggests a key failure, false otherwise.
|
| 105 |
*/
|
| 106 |
-
function isKeyRelatedError(error) {
|
| 107 |
-
const errorMessage = error?.message?.toLowerCase() || '';
|
| 108 |
-
const errorStatus = error?.status; // Assuming the error object might have a status property
|
| 109 |
-
|
| 110 |
-
// Check for common indicators of key issues
|
| 111 |
-
if (errorStatus === 401 || errorStatus === 403 || // Unauthorized, Forbidden
|
| 112 |
-
errorMessage.includes('authentication failed') ||
|
| 113 |
-
errorMessage.includes('invalid api key') ||
|
| 114 |
-
errorMessage.includes('permission denied')) {
|
| 115 |
-
return true;
|
| 116 |
-
}
|
| 117 |
-
if (errorStatus === 429 || // Too Many Requests (Rate Limit / Quota)
|
| 118 |
-
errorMessage.includes('rate limit exceeded') ||
|
| 119 |
errorMessage.includes('quota exceeded')) {
|
| 120 |
return true;
|
| 121 |
}
|
|
@@ -124,6 +125,38 @@ function isKeyRelatedError(error) {
|
|
| 124 |
return false;
|
| 125 |
}
|
| 126 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 127 |
// 定义 fal-ai/any-llm 支持的模型列表 (Remains the same)
|
| 128 |
const FAL_SUPPORTED_MODELS = [
|
| 129 |
"anthropic/claude-3.7-sonnet",
|
|
@@ -156,337 +189,197 @@ const getOwner = (modelId) => {
|
|
| 156 |
// GET /v1/models endpoint (Remains the same)
|
| 157 |
app.get('/v1/models', (req, res) => {
|
| 158 |
console.log("Received request for GET /v1/models");
|
| 159 |
-
|
| 160 |
-
|
| 161 |
-
id: modelId, object: "model", created: 1700000000, owned_by: getOwner(modelId)
|
| 162 |
-
}));
|
| 163 |
-
res.json({ object: "list", data: modelsData });
|
| 164 |
-
console.log("Successfully returned model list.");
|
| 165 |
-
} catch (error) {
|
| 166 |
-
console.error("Error processing GET /v1/models:", error);
|
| 167 |
-
res.status(500).json({ error: "Failed to retrieve model list." });
|
| 168 |
-
}
|
| 169 |
-
});
|
| 170 |
-
|
| 171 |
-
// === convertMessagesToFalPrompt 函数 (Remains the same) ===
|
| 172 |
-
function convertMessagesToFalPrompt(messages) {
|
| 173 |
-
let fixed_system_prompt_content = "";
|
| 174 |
-
const conversation_message_blocks = [];
|
| 175 |
-
// console.log(`Original messages count: ${messages.length}`); // Less verbose logging
|
| 176 |
-
|
| 177 |
-
// 1. 分离 System 消息,格式化 User/Assistant 消息
|
| 178 |
-
for (const message of messages) {
|
| 179 |
-
let content = (message.content === null || message.content === undefined) ? "" : String(message.content);
|
| 180 |
-
switch (message.role) {
|
| 181 |
-
case 'system':
|
| 182 |
-
fixed_system_prompt_content += `System: ${content}\n\n`;
|
| 183 |
-
break;
|
| 184 |
-
case 'user':
|
| 185 |
-
conversation_message_blocks.push(`Human: ${content}\n\n`);
|
| 186 |
-
break;
|
| 187 |
-
case 'assistant':
|
| 188 |
-
conversation_message_blocks.push(`Assistant: ${content}\n\n`);
|
| 189 |
-
break;
|
| 190 |
-
default:
|
| 191 |
-
console.warn(`Unsupported role: ${message.role}`);
|
| 192 |
-
continue;
|
| 193 |
-
}
|
| 194 |
-
}
|
| 195 |
|
| 196 |
-
// 2. 截断合并后的 system 消息(如果超长)
|
| 197 |
-
if (fixed_system_prompt_content.length > SYSTEM_PROMPT_LIMIT) {
|
| 198 |
-
const originalLength = fixed_system_prompt_content.length;
|
| 199 |
-
fixed_system_prompt_content = fixed_system_prompt_content.substring(0, SYSTEM_PROMPT_LIMIT);
|
| 200 |
-
console.warn(`Combined system messages truncated from ${originalLength} to ${SYSTEM_PROMPT_LIMIT}`);
|
| 201 |
-
}
|
| 202 |
-
fixed_system_prompt_content = fixed_system_prompt_content.trim();
|
| 203 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 204 |
|
| 205 |
-
|
| 206 |
-
|
| 207 |
-
|
| 208 |
-
|
| 209 |
-
|
| 210 |
-
|
| 211 |
-
|
| 212 |
-
|
| 213 |
-
|
| 214 |
-
// 4. 反向填充 User/Assistant 对话历史
|
| 215 |
-
const prompt_history_blocks = [];
|
| 216 |
-
const system_prompt_history_blocks = [];
|
| 217 |
-
let current_prompt_length = 0;
|
| 218 |
-
let current_system_history_length = 0;
|
| 219 |
-
let promptFull = false;
|
| 220 |
-
let systemHistoryFull = (remaining_system_limit <= 0);
|
| 221 |
-
|
| 222 |
-
// console.log(`Processing ${conversation_message_blocks.length} user/assistant messages for recency filling.`);
|
| 223 |
-
for (let i = conversation_message_blocks.length - 1; i >= 0; i--) {
|
| 224 |
-
const message_block = conversation_message_blocks[i];
|
| 225 |
-
const block_length = message_block.length;
|
| 226 |
-
|
| 227 |
-
if (promptFull && systemHistoryFull) {
|
| 228 |
-
// console.log(`Both prompt and system history slots full. Omitting older messages from index ${i}.`);
|
| 229 |
-
break;
|
| 230 |
}
|
| 231 |
|
| 232 |
-
|
| 233 |
-
|
| 234 |
-
|
| 235 |
-
|
| 236 |
-
|
| 237 |
-
|
| 238 |
-
|
| 239 |
-
|
| 240 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 241 |
}
|
| 242 |
-
}
|
|
|
|
|
|
|
| 243 |
|
| 244 |
-
|
| 245 |
-
|
| 246 |
-
|
| 247 |
-
system_prompt_history_blocks.unshift(message_block);
|
| 248 |
-
current_system_history_length += block_length;
|
| 249 |
-
continue;
|
| 250 |
} else {
|
| 251 |
-
|
| 252 |
-
|
| 253 |
}
|
| 254 |
-
}
|
| 255 |
-
}
|
| 256 |
|
| 257 |
-
|
| 258 |
-
const system_prompt_history_content = system_prompt_history_blocks.join('').trim();
|
| 259 |
-
const final_prompt = prompt_history_blocks.join('').trim();
|
| 260 |
|
| 261 |
-
// 定义分隔符
|
| 262 |
-
const SEPARATOR = "\n\n-------下面是比较早之前的对话内容-----\n\n";
|
| 263 |
|
| 264 |
-
let final_system_prompt = "";
|
| 265 |
|
| 266 |
-
const hasFixedSystem = fixed_system_prompt_content.length > 0;
|
| 267 |
-
const hasSystemHistory = system_prompt_history_content.length > 0;
|
| 268 |
|
| 269 |
-
if (hasFixedSystem && hasSystemHistory) {
|
| 270 |
-
final_system_prompt = fixed_system_prompt_content + SEPARATOR + system_prompt_history_content;
|
| 271 |
-
// console.log("Combining fixed system prompt and history with separator.");
|
| 272 |
-
} else if (hasFixedSystem) {
|
| 273 |
-
final_system_prompt = fixed_system_prompt_content;
|
| 274 |
-
// console.log("Using only fixed system prompt.");
|
| 275 |
-
} else if (hasSystemHistory) {
|
| 276 |
-
final_system_prompt = system_prompt_history_content;
|
| 277 |
-
// console.log("Using only history in system prompt slot.");
|
| 278 |
-
}
|
| 279 |
|
| 280 |
-
// 6. 返回结果
|
| 281 |
-
const result = {
|
| 282 |
-
system_prompt: final_system_prompt,
|
| 283 |
-
prompt: final_prompt
|
| 284 |
-
};
|
| 285 |
|
| 286 |
-
console.log(`Final system_prompt length: ${result.system_prompt.length}, Final prompt length: ${result.prompt.length}`);
|
| 287 |
|
| 288 |
-
return result;
|
| 289 |
-
}
|
| 290 |
-
// === convertMessagesToFalPrompt 函数结束 ===
|
| 291 |
|
| 292 |
-
/**
|
| 293 |
-
* 使用当前请求的FAL key配置fal客户端
|
| 294 |
-
* @param {object} req - Express请求对象
|
| 295 |
-
*/
|
| 296 |
-
function configureFalClientForRequest(req) {
|
| 297 |
-
const falKey = requestContext.get(req);
|
| 298 |
-
if (!falKey) {
|
| 299 |
-
throw new Error('未找到FAL API Key');
|
| 300 |
}
|
| 301 |
-
fal.config({ credentials: falKey });
|
| 302 |
-
}
|
| 303 |
|
| 304 |
-
|
| 305 |
-
|
| 306 |
-
* @param {'stream' | 'subscribe'} operation - The fal operation to perform.
|
| 307 |
-
* @param {string} functionId - The fal function ID (e.g., "fal-ai/any-llm").
|
| 308 |
-
* @param {object} params - The parameters for the fal function call (input, logs, etc.).
|
| 309 |
-
* @returns {Promise<any>} - The result from the successful fal call (stream or subscription result).
|
| 310 |
-
* @throws {Error} - Throws an error if all keys fail or a non-key-related error occurs.
|
| 311 |
-
*/
|
| 312 |
-
async function tryFalCallWithFailover(req, operation, functionId, params) {
|
| 313 |
-
try {
|
| 314 |
-
configureFalClientForRequest(req);
|
| 315 |
-
|
| 316 |
-
if (operation === 'stream') {
|
| 317 |
-
const streamResult = await fal.stream(functionId, params);
|
| 318 |
-
return streamResult;
|
| 319 |
-
} else {
|
| 320 |
-
const result = await fal.subscribe(functionId, params);
|
| 321 |
-
return result;
|
| 322 |
-
}
|
| 323 |
-
} catch (error) {
|
| 324 |
-
console.error('FAL API调用错误:', error.message || error);
|
| 325 |
-
throw error;
|
| 326 |
-
} finally {
|
| 327 |
-
// 清理请求上下文
|
| 328 |
-
requestContext.delete(req);
|
| 329 |
-
}
|
| 330 |
}
|
| 331 |
|
|
|
|
| 332 |
// POST /v1/chat/completions endpoint (Modified to use tryFalCallWithFailover)
|
| 333 |
app.post('/v1/chat/completions', async (req, res) => {
|
| 334 |
const { model, messages, stream = false, reasoning = false, ...restOpenAIParams } = req.body;
|
| 335 |
|
| 336 |
-
|
|
|
|
| 337 |
|
| 338 |
if (!FAL_SUPPORTED_MODELS.includes(model)) {
|
| 339 |
-
|
| 340 |
}
|
| 341 |
if (!model || !messages || !Array.isArray(messages) || messages.length === 0) {
|
| 342 |
-
console.error("
|
| 343 |
-
return res.status(400).json({ error: '缺少或无效的参数: 需要model和messages数组。' });
|
| 344 |
-
}
|
| 345 |
-
|
| 346 |
-
try {
|
| 347 |
-
const { prompt, system_prompt } = convertMessagesToFalPrompt(messages);
|
| 348 |
-
|
| 349 |
-
const falInput = {
|
| 350 |
-
model: model,
|
| 351 |
-
prompt: prompt,
|
| 352 |
-
...(system_prompt && { system_prompt: system_prompt }),
|
| 353 |
-
reasoning: !!reasoning,
|
| 354 |
-
};
|
| 355 |
-
|
| 356 |
-
console.log("Prepared Fal Input (lengths):", { system_prompt: system_prompt?.length, prompt: prompt?.length });
|
| 357 |
-
|
| 358 |
-
if (stream) {
|
| 359 |
-
res.setHeader('Content-Type', 'text/event-stream; charset=utf-8');
|
| 360 |
-
res.setHeader('Cache-Control', 'no-cache');
|
| 361 |
-
res.setHeader('Connection', 'keep-alive');
|
| 362 |
-
res.setHeader('Access-Control-Allow-Origin', '*');
|
| 363 |
-
res.flushHeaders();
|
| 364 |
-
|
| 365 |
-
let previousOutput = '';
|
| 366 |
let falStream;
|
| 367 |
|
| 368 |
try {
|
| 369 |
-
|
| 370 |
|
| 371 |
-
|
| 372 |
const currentOutput = (event && typeof event.output === 'string') ? event.output : '';
|
| 373 |
const isPartial = (event && typeof event.partial === 'boolean') ? event.partial : true;
|
| 374 |
const errorInfo = (event && event.error) ? event.error : null;
|
| 375 |
-
|
| 376 |
-
if (errorInfo) {
|
| 377 |
-
console.error("Error received *during* fal stream:", errorInfo);
|
| 378 |
-
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)}` } }] };
|
| 379 |
-
res.write(`data: ${JSON.stringify(errorChunk)}\n\n`);
|
| 380 |
-
break;
|
| 381 |
-
}
|
| 382 |
-
|
| 383 |
-
let deltaContent = '';
|
| 384 |
-
if (currentOutput.startsWith(previousOutput)) {
|
| 385 |
-
deltaContent = currentOutput.substring(previousOutput.length);
|
| 386 |
-
} else if (currentOutput.length > 0) {
|
| 387 |
-
console.warn("Fal stream output mismatch detected. Sending full current output as delta.", { previousLength: previousOutput.length, currentLength: currentOutput.length });
|
| 388 |
-
deltaContent = currentOutput;
|
| 389 |
-
previousOutput = '';
|
| 390 |
-
}
|
| 391 |
-
previousOutput = currentOutput;
|
| 392 |
-
|
| 393 |
-
if (deltaContent || !isPartial) {
|
| 394 |
const openAIChunk = { id: `chatcmpl-${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 }] };
|
| 395 |
res.write(`data: ${JSON.stringify(openAIChunk)}\n\n`);
|
| 396 |
}
|
| 397 |
-
|
| 398 |
-
|
| 399 |
-
|
| 400 |
-
|
| 401 |
|
| 402 |
} catch (streamError) {
|
| 403 |
-
console.error('
|
| 404 |
if (!res.writableEnded) {
|
| 405 |
-
|
| 406 |
-
|
| 407 |
-
|
| 408 |
-
|
| 409 |
-
|
| 410 |
-
|
| 411 |
-
|
| 412 |
-
|
| 413 |
-
|
| 414 |
-
|
| 415 |
-
|
| 416 |
}
|
| 417 |
|
| 418 |
} else { // Non-stream
|
| 419 |
-
|
| 420 |
-
|
| 421 |
-
|
| 422 |
-
|
| 423 |
-
|
| 424 |
-
|
| 425 |
-
|
| 426 |
-
|
| 427 |
-
|
| 428 |
-
|
| 429 |
-
|
| 430 |
-
|
| 431 |
-
|
| 432 |
-
|
| 433 |
-
|
| 434 |
-
|
| 435 |
-
|
| 436 |
-
|
| 437 |
-
|
| 438 |
-
|
| 439 |
-
|
| 440 |
-
|
| 441 |
-
|
| 442 |
-
|
| 443 |
-
|
| 444 |
-
|
| 445 |
-
|
| 446 |
-
|
| 447 |
-
|
| 448 |
-
|
| 449 |
-
|
| 450 |
-
|
| 451 |
-
|
| 452 |
-
|
| 453 |
-
|
| 454 |
-
|
| 455 |
-
|
| 456 |
-
|
| 457 |
-
|
| 458 |
}
|
| 459 |
|
| 460 |
} catch (error) {
|
| 461 |
-
console.error('
|
| 462 |
if (!res.headersSent) {
|
| 463 |
const errorMessage = (error instanceof Error) ? error.message : JSON.stringify(error);
|
|
|
|
| 464 |
res.status(500).json({
|
| 465 |
-
|
| 466 |
-
|
| 467 |
-
|
| 468 |
-
|
| 469 |
-
|
| 470 |
});
|
| 471 |
} else if (!res.writableEnded) {
|
| 472 |
-
|
| 473 |
-
|
| 474 |
}
|
| 475 |
}
|
| 476 |
});
|
| 477 |
-
|
| 478 |
// --- Server Start ---
|
| 479 |
app.listen(PORT, () => {
|
| 480 |
console.log(`===========================================================`);
|
| 481 |
-
console.log(` Fal OpenAI
|
| 482 |
-
console.log(`
|
| 483 |
-
|
|
|
|
|
|
|
|
|
|
| 484 |
console.log(` Chat Completions: POST http://localhost:${PORT}/v1/chat/completions`);
|
| 485 |
console.log(` Models Endpoint: GET http://localhost:${PORT}/v1/models`);
|
| 486 |
-
console.log(`===========================================================`);
|
| 487 |
-
});
|
| 488 |
|
| 489 |
// Root path response
|
| 490 |
app.get('/', (req, res) => {
|
| 491 |
-
res.send('Fal OpenAI
|
| 492 |
});
|
|
|
|
| 1 |
import express from 'express';
|
| 2 |
import { fal } from '@fal-ai/client';
|
| 3 |
|
| 4 |
+
// --- Multi-Key Configuration ---
|
| 5 |
+
// *** 使用 FAL_KEY 环境变量读取逗号分隔的密钥 ***
|
| 6 |
+
const rawFalKeys = process.env.FAL_KEY; // Expect comma-separated keys: key1,key2,key3 in FAL_KEY
|
| 7 |
+
const API_KEY = process.env.API_KEY; // Custom API Key for proxy auth remains the same
|
| 8 |
+
|
| 9 |
+
if (!rawFalKeys) {
|
| 10 |
+
// *** 更新错误信息以引用 FAL_KEY ***
|
| 11 |
+
console.error("Error: FAL_KEY environment variable is not set (should be comma-separated).");
|
| 12 |
+
process.exit(1);
|
| 13 |
+
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
|
| 15 |
+
if (!API_KEY) {
|
| 16 |
+
console.error("Error: API_KEY environment variable is not set.");
|
| 17 |
+
process.exit(1);
|
| 18 |
+
}
|
|
|
|
| 19 |
|
| 20 |
+
// Parse and prepare the keys
|
| 21 |
+
let falKeys = rawFalKeys.split(',')
|
| 22 |
+
.map(key => key.trim())
|
| 23 |
+
.filter(key => key.length > 0)
|
| 24 |
+
.map(key => ({
|
| 25 |
+
key: key,
|
| 26 |
+
failed: false, // Track if the key is currently considered failed
|
| 27 |
+
failedTimestamp: 0 // Timestamp when the key was marked as failed
|
| 28 |
+
}));
|
| 29 |
+
|
| 30 |
+
if (falKeys.length === 0) {
|
| 31 |
+
// *** 更新错误信息以引用 FAL_KEY ***
|
| 32 |
+
console.error("Error: No valid keys found in FAL_KEY after processing the environment variable.");
|
| 33 |
+
process.exit(1);
|
| 34 |
+
}
|
| 35 |
|
| 36 |
+
let currentKeyIndex = 0;
|
| 37 |
+
const failedKeyCooldown = 60 * 1000; // Cooldown period in milliseconds (e.g., 60 seconds) before retrying a failed key
|
|
|
|
|
|
|
| 38 |
|
| 39 |
+
// *** 更新日志信息以引用 FAL_KEY ***
|
| 40 |
+
console.log(`Loaded ${falKeys.length} FAL API Key(s) from FAL_KEY environment variable.`);
|
| 41 |
+
console.log(`Failed key cooldown period: ${failedKeyCooldown / 1000} seconds.`);
|
| 42 |
|
| 43 |
// NOTE: We will configure fal client per request now, so initial global config is removed.
|
| 44 |
|
|
|
|
| 80 |
return null; // All keys are currently failed and within cooldown
|
| 81 |
}
|
| 82 |
|
| 83 |
+
|
| 84 |
+
|
| 85 |
+
|
| 86 |
+
|
| 87 |
+
|
| 88 |
+
|
| 89 |
+
|
| 90 |
+
|
| 91 |
+
|
| 92 |
+
|
| 93 |
+
|
| 94 |
+
|
| 95 |
+
|
| 96 |
+
|
| 97 |
+
|
| 98 |
+
|
| 99 |
+
|
| 100 |
/**
|
| 101 |
* Marks a specific key as failed.
|
| 102 |
* @param {object} keyInfo - The key info object to mark as failed.
|
|
|
|
| 110 |
}
|
| 111 |
}
|
| 112 |
|
| 113 |
+
|
| 114 |
/**
|
| 115 |
* Determines if an error likely indicates an API key issue (auth, quota, etc.).
|
| 116 |
* This needs refinement based on actual errors from fal.ai.
|
| 117 |
* @param {Error} error - The error object caught from the fal client.
|
| 118 |
* @returns {boolean} - True if the error suggests a key failure, false otherwise.
|
| 119 |
*/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 120 |
errorMessage.includes('quota exceeded')) {
|
| 121 |
return true;
|
| 122 |
}
|
|
|
|
| 125 |
return false;
|
| 126 |
}
|
| 127 |
|
| 128 |
+
// --- Express App Setup ---
|
| 129 |
+
const app = express();
|
| 130 |
+
app.use(express.json({ limit: '50mb' }));
|
| 131 |
+
app.use(express.urlencoded({ extended: true, limit: '50mb' }));
|
| 132 |
+
|
| 133 |
+
const PORT = process.env.PORT || 3000;
|
| 134 |
+
|
| 135 |
+
// API Key 鉴权中间件 (Remains the same, checks custom API_KEY)
|
| 136 |
+
const apiKeyAuth = (req, res, next) => {
|
| 137 |
+
const authHeader = req.headers['authorization'];
|
| 138 |
+
|
| 139 |
+
}
|
| 140 |
+
|
| 141 |
+
const providedKey = authParts[1];
|
| 142 |
+
if (providedKey !== API_KEY) {
|
| 143 |
+
console.warn('Unauthorized: Invalid API Key');
|
| 144 |
+
return res.status(401).json({ error: 'Unauthorized: Invalid API Key' });
|
| 145 |
+
}
|
| 146 |
+
|
| 147 |
+
|
| 148 |
+
|
| 149 |
+
|
| 150 |
+
next();
|
| 151 |
+
};
|
| 152 |
+
|
| 153 |
+
app.use(['/v1/models', '/v1/chat/completions'], apiKeyAuth);
|
| 154 |
+
|
| 155 |
+
// === 全局定义限制 === (Remains the same)
|
| 156 |
+
const PROMPT_LIMIT = 4800;
|
| 157 |
+
const SYSTEM_PROMPT_LIMIT = 4800;
|
| 158 |
+
// === 限制定义结束 ===
|
| 159 |
+
|
| 160 |
// 定义 fal-ai/any-llm 支持的模型列表 (Remains the same)
|
| 161 |
const FAL_SUPPORTED_MODELS = [
|
| 162 |
"anthropic/claude-3.7-sonnet",
|
|
|
|
| 189 |
// GET /v1/models endpoint (Remains the same)
|
| 190 |
app.get('/v1/models', (req, res) => {
|
| 191 |
console.log("Received request for GET /v1/models");
|
| 192 |
+
}
|
| 193 |
+
// === convertMessagesToFalPrompt 函数结束 ===
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 194 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 195 |
|
| 196 |
+
/**
|
| 197 |
+
* Wraps the fal.ai API call with retry logic using available keys.
|
| 198 |
+
* @param {'stream' | 'subscribe'} operation - The fal operation to perform.
|
| 199 |
+
* @param {string} functionId - The fal function ID (e.g., "fal-ai/any-llm").
|
| 200 |
+
* @param {object} params - The parameters for the fal function call (input, logs, etc.).
|
| 201 |
+
* @returns {Promise<any>} - The result from the successful fal call (stream or subscription result).
|
| 202 |
+
* @throws {Error} - Throws an error if all keys fail or a non-key-related error occurs.
|
| 203 |
|
| 204 |
+
*/
|
| 205 |
+
async function tryFalCallWithFailover(operation, functionId, params) {
|
| 206 |
+
const maxRetries = falKeys.length; // Try each key at most once per request cycle
|
| 207 |
+
let lastError = null;
|
| 208 |
+
|
| 209 |
+
for (let i = 0; i < maxRetries; i++) {
|
| 210 |
+
const keyInfo = getNextKey();
|
| 211 |
+
if (!keyInfo) {
|
| 212 |
+
throw new Error(lastError ? `All FAL keys failed. Last error: ${lastError.message}` : "All FAL keys are currently unavailable (failed or in cooldown).");
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 213 |
}
|
| 214 |
|
| 215 |
+
const currentFalKey = keyInfo.key;
|
| 216 |
+
console.log(`Attempt ${i + 1}/${maxRetries}: Using key ending in ...${currentFalKey.slice(-4)}`);
|
| 217 |
+
|
| 218 |
+
try {
|
| 219 |
+
// --- Configure fal client with the selected key for this attempt ---
|
| 220 |
+
// WARNING: This global config change might have concurrency issues in high-load scenarios
|
| 221 |
+
// if the fal client library doesn't isolate requests properly.
|
| 222 |
+
fal.config({ credentials: currentFalKey });
|
| 223 |
+
|
| 224 |
+
if (operation === 'stream') {
|
| 225 |
+
const streamResult = await fal.stream(functionId, params);
|
| 226 |
+
console.log(`Successfully initiated stream with key ending in ...${currentFalKey.slice(-4)}`);
|
| 227 |
+
return streamResult;
|
| 228 |
+
} else { // 'subscribe' (non-stream)
|
| 229 |
+
const result = await fal.subscribe(functionId, params);
|
| 230 |
+
console.log(`Successfully completed subscribe request with key ending in ...${currentFalKey.slice(-4)}`);
|
| 231 |
+
|
| 232 |
+
if (result && result.error) {
|
| 233 |
+
console.warn(`Fal-ai returned an application error (non-stream) with key ...${currentFalKey.slice(-4)}: ${JSON.stringify(result.error)}`);
|
| 234 |
+
}
|
| 235 |
+
return result;
|
| 236 |
}
|
| 237 |
+
} catch (error) {
|
| 238 |
+
console.error(`Error using key ending in ...${currentFalKey.slice(-4)}:`, error.message || error);
|
| 239 |
+
lastError = error;
|
| 240 |
|
| 241 |
+
if (isKeyRelatedError(error)) {
|
| 242 |
+
markKeyFailed(keyInfo);
|
| 243 |
+
console.log(`Key marked as failed. Trying next key if available...`);
|
|
|
|
|
|
|
|
|
|
| 244 |
} else {
|
| 245 |
+
console.error("Non-key related error occurred. Aborting retries.");
|
| 246 |
+
throw error;
|
| 247 |
}
|
|
|
|
|
|
|
| 248 |
|
| 249 |
+
}
|
|
|
|
|
|
|
| 250 |
|
|
|
|
|
|
|
| 251 |
|
|
|
|
| 252 |
|
|
|
|
|
|
|
| 253 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 254 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 255 |
|
|
|
|
| 256 |
|
|
|
|
|
|
|
|
|
|
| 257 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 258 |
}
|
|
|
|
|
|
|
| 259 |
|
| 260 |
+
console.error("All FAL keys failed after attempting each one.");
|
| 261 |
+
throw new Error(lastError ? `All FAL keys failed. Last error: ${lastError.message}` : "All FAL API keys failed.");
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 262 |
}
|
| 263 |
|
| 264 |
+
|
| 265 |
// POST /v1/chat/completions endpoint (Modified to use tryFalCallWithFailover)
|
| 266 |
app.post('/v1/chat/completions', async (req, res) => {
|
| 267 |
const { model, messages, stream = false, reasoning = false, ...restOpenAIParams } = req.body;
|
| 268 |
|
| 269 |
+
|
| 270 |
+
console.log(`Received chat completion request for model: ${model}, stream: ${stream}`);
|
| 271 |
|
| 272 |
if (!FAL_SUPPORTED_MODELS.includes(model)) {
|
| 273 |
+
console.warn(`Warning: Requested model '${model}' is not in the explicitly supported list.`);
|
| 274 |
}
|
| 275 |
if (!model || !messages || !Array.isArray(messages) || messages.length === 0) {
|
| 276 |
+
console.error("Invalid request parameters:", { model, messages: Array.isArray(messages) ? messages.length : typeof messages });
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 277 |
let falStream;
|
| 278 |
|
| 279 |
try {
|
| 280 |
+
falStream = await tryFalCallWithFailover('stream', "fal-ai/any-llm", { input: falInput });
|
| 281 |
|
| 282 |
+
for await (const event of falStream) {
|
| 283 |
const currentOutput = (event && typeof event.output === 'string') ? event.output : '';
|
| 284 |
const isPartial = (event && typeof event.partial === 'boolean') ? event.partial : true;
|
| 285 |
const errorInfo = (event && event.error) ? event.error : null;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 286 |
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 }] };
|
| 287 |
res.write(`data: ${JSON.stringify(openAIChunk)}\n\n`);
|
| 288 |
}
|
| 289 |
+
}
|
| 290 |
+
res.write(`data: [DONE]\n\n`);
|
| 291 |
+
res.end();
|
| 292 |
+
console.log("Stream finished successfully.");
|
| 293 |
|
| 294 |
} catch (streamError) {
|
| 295 |
+
console.error('Error during stream processing:', streamError);
|
| 296 |
if (!res.writableEnded) {
|
| 297 |
+
try {
|
| 298 |
+
const errorDetails = (streamError instanceof Error) ? streamError.message : JSON.stringify(streamError);
|
| 299 |
+
const finalErrorChunk = { error: { message: "Stream failed", type: "proxy_error", details: errorDetails } };
|
| 300 |
+
res.write(`data: ${JSON.stringify(finalErrorChunk)}\n\n`);
|
| 301 |
+
res.write(`data: [DONE]\n\n`);
|
| 302 |
+
res.end();
|
| 303 |
+
} catch (finalError) {
|
| 304 |
+
console.error('Error sending final stream error message to client:', finalError);
|
| 305 |
+
if (!res.writableEnded) { res.end(); }
|
| 306 |
+
}
|
| 307 |
+
}
|
| 308 |
}
|
| 309 |
|
| 310 |
} else { // Non-stream
|
| 311 |
+
console.log("Executing non-stream request with failover...");
|
| 312 |
+
const result = await tryFalCallWithFailover('subscribe', "fal-ai/any-llm", { input: falInput, logs: true });
|
| 313 |
+
|
| 314 |
+
console.log("Received non-stream result from fal-ai via failover wrapper.");
|
| 315 |
+
|
| 316 |
+
if (result && result.error) {
|
| 317 |
+
console.error("Fal-ai returned an application error in non-stream mode (after successful API call):", result.error);
|
| 318 |
+
return res.status(500).json({
|
| 319 |
+
object: "error",
|
| 320 |
+
message: `Fal-ai application error: ${JSON.stringify(result.error)}`,
|
| 321 |
+
type: "fal_ai_error",
|
| 322 |
+
param: null,
|
| 323 |
+
code: result.error.code || null
|
| 324 |
+
});
|
| 325 |
+
}
|
| 326 |
+
|
| 327 |
+
const openAIResponse = {
|
| 328 |
+
id: `chatcmpl-${result?.requestId || Date.now()}`,
|
| 329 |
+
object: "chat.completion",
|
| 330 |
+
created: Math.floor(Date.now() / 1000),
|
| 331 |
+
model: model,
|
| 332 |
+
choices: [{
|
| 333 |
+
index: 0,
|
| 334 |
+
message: {
|
| 335 |
+
role: "assistant",
|
| 336 |
+
content: result?.output || ""
|
| 337 |
+
},
|
| 338 |
+
finish_reason: "stop"
|
| 339 |
+
}],
|
| 340 |
+
usage: {
|
| 341 |
+
prompt_tokens: null,
|
| 342 |
+
completion_tokens: null,
|
| 343 |
+
total_tokens: null
|
| 344 |
+
},
|
| 345 |
+
system_fingerprint: null,
|
| 346 |
+
...(result?.reasoning && { fal_reasoning: result.reasoning }),
|
| 347 |
+
};
|
| 348 |
+
res.json(openAIResponse);
|
| 349 |
+
console.log("Returned non-stream response successfully.");
|
| 350 |
}
|
| 351 |
|
| 352 |
} catch (error) {
|
| 353 |
+
console.error('Unhandled error in /v1/chat/completions:', error);
|
| 354 |
if (!res.headersSent) {
|
| 355 |
const errorMessage = (error instanceof Error) ? error.message : JSON.stringify(error);
|
| 356 |
+
const errorType = error.message?.includes("All FAL keys failed") ? "api_key_error" : "proxy_internal_error";
|
| 357 |
res.status(500).json({
|
| 358 |
+
error: {
|
| 359 |
+
message: `Internal Server Error in Proxy: ${errorMessage}`,
|
| 360 |
+
type: errorType,
|
| 361 |
+
details: error.stack // Optional: include stack in dev/debug mode
|
| 362 |
+
}
|
| 363 |
});
|
| 364 |
} else if (!res.writableEnded) {
|
| 365 |
+
console.error("Headers already sent, attempting to end response after error.");
|
| 366 |
+
res.end();
|
| 367 |
}
|
| 368 |
}
|
| 369 |
});
|
|
|
|
| 370 |
// --- Server Start ---
|
| 371 |
app.listen(PORT, () => {
|
| 372 |
console.log(`===========================================================`);
|
| 373 |
+
console.log(` Fal OpenAI Proxy Server (Multi-Key Failover)`);
|
| 374 |
+
console.log(` Listening on port: ${PORT}`);
|
| 375 |
+
// *** 更新日志信息以引用 FAL_KEY ***
|
| 376 |
+
console.log(` Loaded ${falKeys.length} FAL API Key(s) from FAL_KEY.`);
|
| 377 |
+
console.log(` API Key Auth Enabled: ${API_KEY ? 'Yes' : 'No'}`);
|
| 378 |
+
console.log(` Limits: System Prompt=${SYSTEM_PROMPT_LIMIT}, Prompt=${PROMPT_LIMIT}`);
|
| 379 |
console.log(` Chat Completions: POST http://localhost:${PORT}/v1/chat/completions`);
|
| 380 |
console.log(` Models Endpoint: GET http://localhost:${PORT}/v1/models`);
|
|
|
|
|
|
|
| 381 |
|
| 382 |
// Root path response
|
| 383 |
app.get('/', (req, res) => {
|
| 384 |
+
res.send('Fal OpenAI Proxy (Multi-Key Failover) is running.');
|
| 385 |
});
|