feat: update LLM error handling with retry logic and enhance model validation messages

server/src/services/llm.ts

@@ -1,8 +1,141 @@
-import OpenAI from 'openai';
+import OpenAI, { APIError, RateLimitError, APIConnectionError, AuthenticationError } from 'openai';
 import type { ChatCompletionContentPart } from 'openai/resources/chat/completions';
 
 interface LLMConfig { apiKey: string; baseURL: string; model: string; }
 
+// 重试配置
+const RETRY_CONFIG = {
+  maxRetries: 3,
+  initialDelayMs: 1000,
+  maxDelayMs: 30000,
+};
+
+// 自定义错误类型，用于前端识别
+export class LLMError extends Error {
+  constructor(
+    message: string,
+    public readonly code: string,
+    public readonly statusCode?: number,
+    public readonly retryable: boolean = false
+  ) {
+    super(message);
+    this.name = 'LLMError';
+  }
+}
+
+const sleep = (ms: number) => new Promise(resolve => setTimeout(resolve, ms));
+
+function handleAPIError(error: unknown): never {
+  if (error instanceof RateLimitError) {
+    throw new LLMError(
+      'API rate limit exceeded. Please wait a moment and try again.',
+      'RATE_LIMIT',
+      429,
+      true
+    );
+  }
+
+  if (error instanceof AuthenticationError) {
+    throw new LLMError(
+      'Invalid API key. Please check your API key configuration.',
+      'AUTH_ERROR',
+      401,
+      false
+    );
+  }
+
+  if (error instanceof APIConnectionError) {
+    throw new LLMError(
+      'Failed to connect to API server. Please check your network or API base URL.',
+      'CONNECTION_ERROR',
+      undefined,
+      true
+    );
+  }
+
+  if (error instanceof APIError) {
+    const statusCode = error.status;
+    if (statusCode === 429) {
+      throw new LLMError(
+        'API rate limit exceeded. Please wait a moment and try again.',
+        'RATE_LIMIT',
+        429,
+        true
+      );
+    }
+    if (statusCode === 401 || statusCode === 403) {
+      throw new LLMError(
+        'Authentication failed. Please check your API key.',
+        'AUTH_ERROR',
+        statusCode,
+        false
+      );
+    }
+    if (statusCode === 400) {
+      throw new LLMError(
+        'Invalid request. Please check your input parameters.',
+        'BAD_REQUEST',
+        400,
+        false
+      );
+    }
+    if (statusCode && statusCode >= 500) {
+      throw new LLMError(
+        'API server error. Please try again later.',
+        'SERVER_ERROR',
+        statusCode,
+        true
+      );
+    }
+    throw new LLMError(
+      error.message || 'An unexpected API error occurred.',
+      'API_ERROR',
+      statusCode,
+      false
+    );
+  }
+
+  if (error instanceof Error) {
+    throw new LLMError(error.message, 'UNKNOWN_ERROR', undefined, false);
+  }
+
+  throw new LLMError('An unexpected error occurred.', 'UNKNOWN_ERROR', undefined, false);
+}
+
+async function withRetry<T>(
+  fn: () => Promise<T>,
+  maxRetries = RETRY_CONFIG.maxRetries
+): Promise<T> {
+  let lastError: Error = new Error('Unknown error');
+  let delay = RETRY_CONFIG.initialDelayMs;
+
+  for (let attempt = 0; attempt <= maxRetries; attempt++) {
+    try {
+      return await fn();
+    } catch (error) {
+      lastError = error as Error;
+
+      // 检查是否是可重试的错误
+      const isRetryable =
+        error instanceof RateLimitError ||
+        error instanceof APIConnectionError ||
+        (error instanceof APIError && error.status && error.status >= 500) ||
+        (error instanceof LLMError && error.retryable);
+
+      if (isRetryable && attempt < maxRetries) {
+        console.log(`Retry attempt ${attempt + 1}/${maxRetries} after ${delay}ms...`);
+        await sleep(delay);
+        delay = Math.min(delay * 2, RETRY_CONFIG.maxDelayMs);
+        continue;
+      }
+
+      handleAPIError(error);
+    }
+  }
+
+  handleAPIError(lastError);
+}
+
 function normalizeBaseURL(url: string): string {
   const parsed = new URL(url.trim());
   if (parsed.protocol !== 'http:' && parsed.protocol !== 'https:') throw new Error('Base URL must use http or https');
@@ -30,22 +163,33 @@ const buildUserContent = (userPrompt: string, pdfDataUrl?: string): MessageConte
 
 export async function generateWithLLM(systemPrompt: string, userPrompt: string, pdfDataUrl?: string, config?: Partial<LLMConfig>): Promise<string> {
   const { client, model } = createClient(config);
-  const response = await client.chat.completions.create({
-    model,
-    messages: [{ role: 'system', content: systemPrompt }, { role: 'user', content: buildUserContent(userPrompt, pdfDataUrl) }],
+
+  return withRetry(async () => {
+    const response = await client.chat.completions.create({
+      model,
+      messages: [{ role: 'system', content: systemPrompt }, { role: 'user', content: buildUserContent(userPrompt, pdfDataUrl) }],
+    });
+    return response.choices[0]?.message?.content || '';
   });
-  return response.choices[0]?.message?.content || '';
 }
 
 export async function* generateWithLLMStream(systemPrompt: string, userPrompt: string, pdfDataUrl?: string, config?: Partial<LLMConfig>): AsyncGenerator<string> {
   const { client, model } = createClient(config);
-  const stream = await client.chat.completions.create({
-    model,
-    messages: [{ role: 'system', content: systemPrompt }, { role: 'user', content: buildUserContent(userPrompt, pdfDataUrl) }],
-    stream: true,
+
+  const stream = await withRetry(async () => {
+    return client.chat.completions.create({
+      model,
+      messages: [{ role: 'system', content: systemPrompt }, { role: 'user', content: buildUserContent(userPrompt, pdfDataUrl) }],
+      stream: true,
+    });
   });
-  for await (const chunk of stream) {
-    const content = chunk.choices[0]?.delta?.content;
-    if (content) yield content;
+
+  try {
+    for await (const chunk of stream) {
+      const content = chunk.choices[0]?.delta?.content;
+      if (content) yield content;
+    }
+  } catch (error) {
+    handleAPIError(error);
   }
 }