feat: OCR 并发识别控制，支持多用户同时上传

- config.py: 新增 OCR_CONCURRENCY 参数（默认1，范围1-10）
- images.py: 用 asyncio.Semaphore 控制 OCR 并发，4个上传端点统一受控
- settings API: 新增 ocr_concurrency 字段读写
- 前端设置页: 新增 OCR 并发数量输入框
- .env / .env.example: 新增 OCR_CONCURRENCY 配置项
- Semaphore 动态重建：修改配置后立即生效，无需重启

.env.example

@@ -33,6 +33,7 @@ LOCAL_BGE_MODEL_PATH=
 # 可选值: paddleocr / aliyun / tencent / deepseek / auto
 # auto 模式: 本地 PaddleOCR 优先，失败后 fallback 到 DeepSeek / 云端 OCR
 OCR_PROVIDER=deepseek
+OCR_CONCURRENCY=1
 
 # ── 阿里云 OCR 配置 ──
 ALIYUN_OCR_ACCESS_KEY=your-aliyun-access-key

app/api/v1/images.py

@@ -30,6 +30,19 @@ from app.services.llm_service import LLMService
 logger = logging.getLogger(__name__)
 router = APIRouter()
 
+# OCR 并发控制信号量（动态读取配置）
+_ocr_semaphore: asyncio.Semaphore | None = None
+
+
+def _get_ocr_semaphore() -> asyncio.Semaphore:
+    """获取/重建 OCR 并发信号量"""
+    global _ocr_semaphore
+    from app.config import settings
+    concurrency = settings.OCR_CONCURRENCY
+    if _ocr_semaphore is None or _ocr_semaphore._value != concurrency:
+        _ocr_semaphore = asyncio.Semaphore(concurrency)
+    return _ocr_semaphore
+
 ALLOWED_EXTENSIONS = {".png", ".jpg", ".jpeg", ".bmp", ".webp"}
 
 # 全局 BM25 索引实例（在 main.py 中初始化）
@@ -126,7 +139,8 @@ async def recognize_image(image_id: int, db: AsyncSession = Depends(get_db)):
     await db.flush()
 
     try:
-        ocr_result = await OCRService.recognize(image_record.file_path)
+        async with _get_ocr_semaphore():
+            ocr_result = await OCRService.recognize(image_record.file_path)
         _save_ocr_result(image_record, ocr_result, db)
         await db.flush()
         await db.refresh(image_record)
@@ -171,7 +185,8 @@ async def batch_recognize(files: List[UploadFile], db: AsyncSession = Depends(ge
         db.add(image_record)
         await db.flush()
         try:
-            ocr_result = await OCRService.recognize(str(dest_path))
+            async with _get_ocr_semaphore():
+                ocr_result = await OCRService.recognize(str(dest_path))
             # OCR 内容去重
             existing_id = await _check_ocr_duplicate(ocr_result.text, db)
             if existing_id is not None:
@@ -239,7 +254,8 @@ async def import_from_paths(data: PathImportRequest, db: AsyncSession = Depends(
         db.add(image_record)
         await db.flush()
         try:
-            ocr_result = await OCRService.recognize(file_path)
+            async with _get_ocr_semaphore():
+                ocr_result = await OCRService.recognize(file_path)
             # OCR 内容去重
             existing_id = await _check_ocr_duplicate(ocr_result.text, db)
             if existing_id is not None:
@@ -279,7 +295,8 @@ async def recognize_direct(file: UploadFile, db: AsyncSession = Depends(get_db))
     db.add(image_record)
     await db.flush()
     try:
-        ocr_result = await OCRService.recognize(str(dest_path))
+        async with _get_ocr_semaphore():
+            ocr_result = await OCRService.recognize(str(dest_path))
         # OCR 内容去重
         existing_id = await _check_ocr_duplicate(ocr_result.text, db)
         if existing_id is not None:

app/api/v1/settings.py

@@ -38,6 +38,7 @@ class SettingsResponse(BaseModel):
     chunk_overlap: int
     search_limit: int
     judge_batch_size: int
+    ocr_concurrency: int
     # API Key 只返回是否已配置（不返回实际值）
     has_minimax_key: bool
     has_deepseek_key: bool
@@ -71,6 +72,7 @@ async def get_settings():
         chunk_overlap=settings.CHUNK_OVERLAP,
         search_limit=settings.SEARCH_LIMIT,
         judge_batch_size=settings.JUDGE_BATCH_SIZE,
+        ocr_concurrency=settings.OCR_CONCURRENCY,
         has_minimax_key=bool(settings.MINIMAX_API_KEY),
         has_deepseek_key=bool(settings.DEEPSEEK_API_KEY),
         has_openai_key=bool(settings.OPENAI_API_KEY),
@@ -110,6 +112,7 @@ class SettingsUpdate(BaseModel):
     chunk_overlap: int | None = None
     search_limit: int | None = None
     judge_batch_size: int | None = None
+    ocr_concurrency: int | None = None
     # 企业微信机器人
     wework_bot_enabled: bool | None = None
     wework_bot_id: str | None = None
@@ -159,6 +162,7 @@ async def update_settings(data: SettingsUpdate):
         "chunk_overlap": "CHUNK_OVERLAP",
         "search_limit": "SEARCH_LIMIT",
         "judge_batch_size": "JUDGE_BATCH_SIZE",
+        "ocr_concurrency": "OCR_CONCURRENCY",
         "wework_bot_enabled": "WEWORK_BOT_ENABLED",
         "wework_bot_id": "WEWORK_BOT_ID",
         "wework_bot_secret": "WEWORK_BOT_SECRET",

app/config.py

@@ -153,6 +153,9 @@ class Settings(BaseSettings):
     SEARCH_LIMIT: int = Field(default=3, description="图片搜索默认返回数量(1-10)")
     JUDGE_BATCH_SIZE: int = Field(default=10, description="每次LLM判断的文章数量(1-50)")
 
+    # ──────────────────────────── OCR ────────────────────────────
+    OCR_CONCURRENCY: int = Field(default=1, ge=1, le=10, description="OCR 并发识别数量(1-10)，多用户同时上传时控制并发")
+
     # ──────────────────────────── CORS ────────────────────────────
     CORS_ORIGINS: list[str] = Field(
         default=["*"],

static/index.html

@@ -1093,6 +1093,14 @@
                                 <span style="color:var(--text-secondary); font-size:13px;">篇（1-50）</span>
                             </div>
                         </div>
+                        <div class="form-group">
+                            <label class="form-label">OCR 并发识别数量</label>
+                            <div style="display:flex; align-items:center; gap:8px;">
+                                <input type="number" id="setting-ocr-concurrency" min="1" max="10" value="1"
+                                       style="width:80px;" class="input-field">
+                                <span style="color:var(--text-secondary); font-size:13px;">张（1-10，多用户同时上传时控制并发）</span>
+                            </div>
+                        </div>
                     </div>
                 </div>
 
@@ -2644,6 +2652,9 @@
             if (data.judge_batch_size != null) {
                 document.getElementById('setting-judge-batch-size').value = data.judge_batch_size;
             }
+            if (data.ocr_concurrency != null) {
+                document.getElementById('setting-ocr-concurrency').value = data.ocr_concurrency;
+            }
             // 企业微信机器人
             document.getElementById('setting-wework-bot-enabled').checked = data.wework_bot_enabled;
             document.getElementById('setting-wework-bot-id').value = data.wework_bot_id || '';
@@ -2673,6 +2684,10 @@
         if (!isNaN(judgeBatchSize) && judgeBatchSize >= 1 && judgeBatchSize <= 50) {
             body.judge_batch_size = judgeBatchSize;
         }
+        const ocrConcurrency = parseInt(document.getElementById('setting-ocr-concurrency').value);
+        if (!isNaN(ocrConcurrency) && ocrConcurrency >= 1 && ocrConcurrency <= 10) {
+            body.ocr_concurrency = ocrConcurrency;
+        }
         // 企业微信机器人
         body.wework_bot_enabled = document.getElementById('setting-wework-bot-enabled').checked;
         const botId = document.getElementById('setting-wework-bot-id').value.trim();