"""
知识库 API
"""
import logging
from typing import Optional, List

from fastapi import APIRouter, Depends, HTTPException, Query, BackgroundTasks
from sqlalchemy import select
from sqlalchemy.ext.asyncio import AsyncSession

from app.database import get_db
from app.base_schema import PaginatedResponse, ResponseModel
from ai_platform.models import LLMModel
from core.application.model import Application
from ai_platform.knowledge.models import KnowledgeBase, KnowledgeDocument
from ai_platform.knowledge.schemas.knowledge_base_schema import (
    KnowledgeBaseCreate,
    KnowledgeBaseUpdate,
    KnowledgeBaseResponse,
    KnowledgeBaseListResponse,
)
from ai_platform.knowledge.schemas.document_schema import (
    DocumentUploadInput,
    DocumentBatchUploadInput,
    DocumentResponse,
    DocumentListResponse,
)
from ai_platform.knowledge.schemas.segment_schema import (
    SegmentResponse,
    SegmentListResponse,
    SegmentUpdateInput,
    SegmentCreateInput,
    RetrievalInput,
    RetrievalResponse,
    ChunkPreviewInput,
    ChunkPreviewResponse,
)
from ai_platform.knowledge.schemas.annotation_schema import (
    AnnotationCreateInput,
    AnnotationUpdateInput,
    AnnotationResponse,
)
from ai_platform.knowledge.models import KnowledgeAnnotation
from ai_platform.knowledge.schemas.retrieval_log_schema import RetrievalLogResponse

logger = logging.getLogger(__name__)

router = APIRouter(prefix="/knowledge", tags=["AI-知识库"])


# ==================== 知识库 CRUD ====================

@router.get("/list", response_model=PaginatedResponse[KnowledgeBaseListResponse], summary="知识库列表")
async def list_knowledge_bases(
    name: Optional[str] = Query(None, description="名称"),
    status: Optional[str] = Query(None, description="状态"),
    application_id: Optional[str] = Query(None, alias="applicationId", description="所属应用ID"),
    page: int = Query(1, ge=1, description="页码"),
    page_size: int = Query(20, ge=1, le=100, alias="pageSize", description="每页数量"),
    db: AsyncSession = Depends(get_db),
):
    """获取知识库列表（自动应用数据权限）"""
    from ai_platform.knowledge.services.knowledge_service import KnowledgeService

    service = KnowledgeService(db)
    items, total = await service.get_list_with_data_scope(
        page=page, page_size=page_size,
        name=name, status=status, application_id=application_id,
    )

    # 批量查询应用名称和模型名称
    app_ids = list({kb.application_id for kb in items if kb.application_id})
    app_name_map = {}
    if app_ids:
        app_result = await db.execute(
            select(Application.id, Application.name).where(Application.id.in_(app_ids))
        )
        app_name_map = {row.id: row.name for row in app_result}

    model_ids = list({kb.embedding_model_id for kb in items if kb.embedding_model_id})
    model_name_map = {}
    if model_ids:
        model_result = await db.execute(
            select(LLMModel.id, LLMModel.display_name).where(LLMModel.id.in_(model_ids))
        )
        model_name_map = {row.id: row.display_name for row in model_result}

    response_items = []
    for kb in items:
        response_items.append({
            "id": kb.id,
            "application_id": kb.application_id,
            "application_name": app_name_map.get(kb.application_id, ""),
            "is_global": kb.is_global or False,
            "name": kb.name,
            "code": kb.code,
            "description": kb.description or "",
            "icon": kb.icon or "",
            "embedding_model_name": model_name_map.get(kb.embedding_model_id, ""),
            "document_count": kb.document_count or 0,
            "segment_count": kb.segment_count or 0,
            "status": kb.status or "active",
            "sys_create_datetime": kb.sys_create_datetime,
        })

    return PaginatedResponse(items=response_items, total=total)


@router.get("/simple", summary="知识库简单列表（下拉选择用）")
async def list_knowledge_bases_simple(
    application_id: Optional[str] = Query(None, alias="applicationId", description="所属应用ID"),
    db: AsyncSession = Depends(get_db),
):
    """获取知识库简单列表"""
    from ai_platform.knowledge.services.knowledge_service import KnowledgeService

    service = KnowledgeService(db)
    return await service.get_simple_list(application_id)


# ==================== 检索日志 ====================

@router.get("/retrieval-logs", response_model=PaginatedResponse[RetrievalLogResponse], summary="检索日志列表")
async def list_retrieval_logs(
    knowledge_base_id: Optional[str] = Query(None, alias="knowledgeBaseId", description="知识库ID筛选"),
    keyword: Optional[str] = Query(None, description="查询关键词搜索"),
    page: int = Query(1, ge=1),
    page_size: int = Query(20, ge=1, le=100, alias="pageSize"),
    db: AsyncSession = Depends(get_db),
):
    """获取检索日志列表"""
    from sqlalchemy import func as sa_func
    from ai_platform.knowledge.models import KnowledgeRetrievalLog

    query = select(KnowledgeRetrievalLog).where(
        KnowledgeRetrievalLog.is_deleted == False,
    )

    if knowledge_base_id:
        from app.db_compat import json_contains
        query = query.where(
            json_contains(KnowledgeRetrievalLog.knowledge_base_ids, knowledge_base_id)
        )
    if keyword:
        query = query.where(KnowledgeRetrievalLog.query.ilike(f"%{keyword}%"))

    count_query = select(sa_func.count()).select_from(query.subquery())
    total = (await db.execute(count_query)).scalar() or 0

    query = query.order_by(KnowledgeRetrievalLog.sys_create_datetime.desc())
    query = query.offset((page - 1) * page_size).limit(page_size)
    result = await db.execute(query)
    items = result.scalars().all()

    return PaginatedResponse(items=items, total=total)


@router.get("/{kb_id}", response_model=KnowledgeBaseResponse, summary="知识库详情")
async def get_knowledge_base(kb_id: str, db: AsyncSession = Depends(get_db)):
    """获取知识库详情"""
    from ai_platform.knowledge.services.knowledge_service import KnowledgeService

    service = KnowledgeService(db)
    kb = await service.get_by_id(kb_id)
    if not kb:
        raise HTTPException(status_code=404, detail="知识库不存在")

    # 获取模型名称
    model_name = ""
    if kb.embedding_model_id:
        model_result = await db.execute(
            select(LLMModel.display_name).where(LLMModel.id == kb.embedding_model_id)
        )
        row = model_result.scalar_one_or_none()
        if row:
            model_name = row

    return _build_kb_response(kb, model_name)


@router.post("", response_model=KnowledgeBaseResponse, summary="创建知识库")
async def create_knowledge_base(data: KnowledgeBaseCreate, db: AsyncSession = Depends(get_db)):
    """创建知识库"""
    from ai_platform.knowledge.services.knowledge_service import KnowledgeService

    service = KnowledgeService(db)
    try:
        kb = await service.create(data)
    except ValueError as e:
        raise HTTPException(status_code=400, detail=str(e))

    return _build_kb_response(kb)


@router.put("/{kb_id}", response_model=KnowledgeBaseResponse, summary="更新知识库")
async def update_knowledge_base(
    kb_id: str,
    data: KnowledgeBaseUpdate,
    db: AsyncSession = Depends(get_db),
):
    """更新知识库"""
    from ai_platform.knowledge.services.knowledge_service import KnowledgeService

    service = KnowledgeService(db)
    kb = await service.update(kb_id, data)
    if not kb:
        raise HTTPException(status_code=404, detail="知识库不存在")

    return _build_kb_response(kb)


@router.delete("/{kb_id}", response_model=ResponseModel, summary="删除知识库")
async def delete_knowledge_base(kb_id: str, db: AsyncSession = Depends(get_db)):
    """删除知识库"""
    from ai_platform.knowledge.services.knowledge_service import KnowledgeService

    service = KnowledgeService(db)
    success = await service.delete(kb_id)
    if not success:
        raise HTTPException(status_code=404, detail="知识库不存在")

    return ResponseModel(message="删除成功")


# ==================== 文档管理 ====================

@router.get("/{kb_id}/documents", response_model=PaginatedResponse[DocumentListResponse], summary="文档列表")
async def list_documents(
    kb_id: str,
    name: Optional[str] = Query(None, description="名称"),
    status: Optional[str] = Query(None, description="状态"),
    page: int = Query(1, ge=1, description="页码"),
    page_size: int = Query(20, ge=1, le=100, alias="pageSize", description="每页数量"),
    db: AsyncSession = Depends(get_db),
):
    """获取知识库文档列表"""
    from ai_platform.knowledge.services.document_service import DocumentService

    service = DocumentService(db)
    items, total = await service.get_list(
        knowledge_base_id=kb_id, page=page, page_size=page_size,
        name=name, status=status,
    )

    response_items = [_build_doc_list_response(doc) for doc in items]
    return PaginatedResponse(items=response_items, total=total)


@router.post("/{kb_id}/documents", response_model=DocumentResponse, summary="添加文档")
async def add_document(
    kb_id: str,
    data: DocumentUploadInput,
    background_tasks: BackgroundTasks,
    db: AsyncSession = Depends(get_db),
):
    """添加文档到知识库（添加后自动开始索引）"""
    from ai_platform.knowledge.services.document_service import DocumentService
    from ai_platform.knowledge.services.indexing_service import IndexingService

    doc_service = DocumentService(db)
    try:
        doc = await doc_service.add_document(kb_id, data.file_id, data.name)
    except ValueError as e:
        raise HTTPException(status_code=400, detail=str(e))

    # 后台异步索引
    background_tasks.add_task(_index_document_task, kb_id, doc.id)

    return _build_doc_response(doc)


@router.post("/{kb_id}/documents/batch", response_model=ResponseModel, summary="批量添加文档")
async def batch_add_documents(
    kb_id: str,
    data: DocumentBatchUploadInput,
    background_tasks: BackgroundTasks,
    db: AsyncSession = Depends(get_db),
):
    """批量添加文档"""
    from ai_platform.knowledge.services.document_service import DocumentService

    doc_service = DocumentService(db)
    docs = await doc_service.batch_add_documents(kb_id, data.file_ids)

    # 后台并发索引所有文档
    doc_ids = [doc.id for doc in docs]
    background_tasks.add_task(_batch_index_documents_task, kb_id, doc_ids)

    return ResponseModel(message=f"成功添加 {len(docs)} 个文档，正在后台索引")


@router.get("/{kb_id}/documents/{doc_id}", response_model=DocumentResponse, summary="文档详情")
async def get_document(kb_id: str, doc_id: str, db: AsyncSession = Depends(get_db)):
    """获取文档详情"""
    from ai_platform.knowledge.services.document_service import DocumentService

    service = DocumentService(db)
    doc = await service.get_by_id(doc_id)
    if not doc or doc.knowledge_base_id != kb_id:
        raise HTTPException(status_code=404, detail="文档不存在")

    return _build_doc_response(doc)


@router.delete("/{kb_id}/documents/{doc_id}", response_model=ResponseModel, summary="删除文档")
async def delete_document(kb_id: str, doc_id: str, db: AsyncSession = Depends(get_db)):
    """删除文档"""
    from ai_platform.knowledge.services.document_service import DocumentService

    service = DocumentService(db)
    success = await service.delete_document(doc_id)
    if not success:
        raise HTTPException(status_code=404, detail="文档不存在")

    return ResponseModel(message="删除成功")


@router.post("/{kb_id}/documents/{doc_id}/reindex", response_model=ResponseModel, summary="重新索引文档")
async def reindex_document(
    kb_id: str,
    doc_id: str,
    background_tasks: BackgroundTasks,
    db: AsyncSession = Depends(get_db),
):
    """重新索引文档"""
    from ai_platform.knowledge.services.document_service import DocumentService

    service = DocumentService(db)
    doc = await service.get_by_id(doc_id)
    if not doc or doc.knowledge_base_id != kb_id:
        raise HTTPException(status_code=404, detail="文档不存在")

    background_tasks.add_task(_index_document_task, kb_id, doc_id)
    return ResponseModel(message="已开始重新索引")


@router.post("/{kb_id}/documents/reindex-all", response_model=ResponseModel, summary="批量重新索引所有文档")
async def reindex_all_documents(
    kb_id: str,
    background_tasks: BackgroundTasks,
    db: AsyncSession = Depends(get_db),
):
    """修改分块策略后一键重新索引知识库所有文档"""
    from ai_platform.knowledge.services.document_service import DocumentService

    service = DocumentService(db)
    docs, total = await service.get_list(kb_id, page=1, page_size=9999, status='completed')
    # 也包含 failed 的文档
    failed_docs, _ = await service.get_list(kb_id, page=1, page_size=9999, status='failed')
    all_docs = docs + failed_docs

    if not all_docs:
        return ResponseModel(message="没有可重新索引的文档")

    doc_ids = [doc.id for doc in all_docs]
    background_tasks.add_task(_batch_index_documents_task, kb_id, doc_ids)

    return ResponseModel(message=f"已开始重新索引 {len(doc_ids)} 个文档")


@router.put("/{kb_id}/documents/{doc_id}/toggle", response_model=DocumentResponse, summary="启用/禁用文档")
async def toggle_document(
    kb_id: str,
    doc_id: str,
    enabled: bool = Query(..., description="是否启用"),
    db: AsyncSession = Depends(get_db),
):
    """启用/禁用文档"""
    from ai_platform.knowledge.services.document_service import DocumentService

    service = DocumentService(db)
    doc = await service.toggle_document(doc_id, enabled)
    if not doc:
        raise HTTPException(status_code=404, detail="文档不存在")

    return _build_doc_response(doc)


# ==================== 分块预览 ====================

@router.post("/{kb_id}/chunk-preview", response_model=ChunkPreviewResponse, summary="分块预览")
async def chunk_preview(
    kb_id: str,
    data: ChunkPreviewInput,
    db: AsyncSession = Depends(get_db),
):
    """预览文档分块结果（不实际入库）"""
    from ai_platform.knowledge.services.knowledge_service import KnowledgeService
    from ai_platform.knowledge.chunking import get_chunker
    from ai_platform.knowledge.services.cleaning_service import CleaningService
    from core.file_manager.service import FileManagerService

    # 验证知识库存在
    service = KnowledgeService(db)
    kb = await service.get_by_id(kb_id)
    if not kb:
        raise HTTPException(status_code=404, detail="知识库不存在")

    # 提取文本
    try:
        text_content = await FileManagerService.get_file_text_content(db, data.file_id)
        if not text_content or not text_content.strip():
            raise HTTPException(status_code=400, detail="文件内容为空")
    except HTTPException:
        raise
    except Exception as e:
        raise HTTPException(status_code=400, detail=f"无法提取文件内容: {str(e)}")

    # 预处理/清洗
    process_rules = data.process_rules or kb.process_rules
    if process_rules:
        text_content = CleaningService.clean(text_content, process_rules)

    # 分块
    strategy = data.chunk_strategy or kb.chunk_strategy or 'recursive'
    chunk_size = data.chunk_size or kb.chunk_size or 500
    chunk_overlap = data.chunk_overlap if data.chunk_overlap is not None else (kb.chunk_overlap or 50)

    # Q&A 模式不支持预览（需要 LLM 调用）
    if strategy == 'qa':
        raise HTTPException(status_code=400, detail="Q&A 分块策略不支持预览")

    chunker = get_chunker(
        strategy=strategy,
        chunk_size=chunk_size,
        chunk_overlap=chunk_overlap,
        separator=data.separator or kb.separator,
    )

    chunks = chunker.chunk(text_content, metadata={})

    # 构建预览结果
    preview_items = []
    for i, chunk in enumerate(chunks):
        char_count = len(chunk.content)
        # 简单 token 估算
        token_count = max(1, int(char_count / 1.5))
        word_count = len(chunk.content.split())
        preview_items.append({
            "position": i,
            "content": chunk.content,
            "char_count": char_count,
            "token_count": token_count,
            "word_count": word_count,
            "answer": chunk.metadata.get('answer'),
            "metadata": chunk.metadata if chunk.metadata else None,
        })

    return ChunkPreviewResponse(
        chunks=preview_items,
        total=len(preview_items),
        strategy=strategy,
        chunk_size=chunk_size,
        chunk_overlap=chunk_overlap,
    )


# ==================== 分段管理 ====================

@router.get("/{kb_id}/segments", response_model=PaginatedResponse[SegmentListResponse], summary="知识库分段列表")
async def list_kb_segments(
    kb_id: str,
    keyword: Optional[str] = Query(None, description="关键词搜索"),
    enabled: Optional[bool] = Query(None, description="是否启用"),
    embedding_status: Optional[str] = Query(None, alias="embeddingStatus", description="向量化状态: completed/failed/pending/skipped"),
    metadata_key: Optional[str] = Query(None, alias="metadataKey", description="元数据key"),
    metadata_value: Optional[str] = Query(None, alias="metadataValue", description="元数据value"),
    page: int = Query(1, ge=1, description="页码"),
    page_size: int = Query(20, ge=1, le=100, alias="pageSize", description="每页数量"),
    db: AsyncSession = Depends(get_db),
):
    """获取知识库所有分段"""
    from ai_platform.knowledge.services.document_service import DocumentService

    service = DocumentService(db)
    items, total = await service.get_kb_segments(
        knowledge_base_id=kb_id, page=page, page_size=page_size,
        keyword=keyword, enabled=enabled, embedding_status=embedding_status,
        metadata_key=metadata_key, metadata_value=metadata_value,
    )

    # 获取文档名称映射
    doc_ids = list({seg.document_id for seg in items})
    doc_name_map = {}
    if doc_ids:
        doc_result = await db.execute(
            select(KnowledgeDocument.id, KnowledgeDocument.name).where(
                KnowledgeDocument.id.in_(doc_ids)
            )
        )
        doc_name_map = {row.id: row.name for row in doc_result}

    response_items = [_build_segment_list_response(seg, doc_name_map) for seg in items]
    return PaginatedResponse(items=response_items, total=total)


@router.get("/{kb_id}/documents/{doc_id}/segments", response_model=PaginatedResponse[SegmentListResponse], summary="文档分段列表")
async def list_document_segments(
    kb_id: str,
    doc_id: str,
    keyword: Optional[str] = Query(None, description="关键词搜索"),
    page: int = Query(1, ge=1, description="页码"),
    page_size: int = Query(20, ge=1, le=100, alias="pageSize", description="每页数量"),
    db: AsyncSession = Depends(get_db),
):
    """获取文档的分段列表"""
    from ai_platform.knowledge.services.document_service import DocumentService

    service = DocumentService(db)
    items, total = await service.get_segments(
        document_id=doc_id, page=page, page_size=page_size, keyword=keyword,
    )

    # 获取文档名称
    doc_result = await db.execute(
        select(KnowledgeDocument.name).where(KnowledgeDocument.id == doc_id)
    )
    doc_name = doc_result.scalar_one_or_none() or ""
    doc_name_map = {doc_id: doc_name}

    response_items = [_build_segment_list_response(seg, doc_name_map) for seg in items]
    return PaginatedResponse(items=response_items, total=total)


@router.put("/{kb_id}/segments/{segment_id}", response_model=SegmentListResponse, summary="更新分段")
async def update_segment(
    kb_id: str,
    segment_id: str,
    data: SegmentUpdateInput,
    db: AsyncSession = Depends(get_db),
):
    """更新分段内容"""
    from ai_platform.knowledge.services.document_service import DocumentService

    service = DocumentService(db)
    segment = await service.update_segment(
        segment_id=segment_id,
        content=data.content,
        keywords=data.keywords,
        enabled=data.enabled,
        extra_metadata=data.extra_metadata,
    )
    if not segment:
        raise HTTPException(status_code=404, detail="分段不存在")

    return _build_segment_list_response(segment, {})


@router.post("/{kb_id}/documents/{doc_id}/segments", response_model=SegmentListResponse, summary="手动添加分段")
async def add_segment(
    kb_id: str,
    doc_id: str,
    data: SegmentCreateInput,
    db: AsyncSession = Depends(get_db),
):
    """手动添加分段"""
    from ai_platform.knowledge.services.document_service import DocumentService

    service = DocumentService(db)
    segment = await service.add_segment(
        knowledge_base_id=kb_id,
        document_id=doc_id,
        content=data.content,
        keywords=data.keywords,
    )

    return _build_segment_list_response(segment, {})


@router.delete("/{kb_id}/segments/{segment_id}", response_model=ResponseModel, summary="删除分段")
async def delete_segment(kb_id: str, segment_id: str, db: AsyncSession = Depends(get_db)):
    """删除分段"""
    from ai_platform.knowledge.services.document_service import DocumentService

    service = DocumentService(db)
    success = await service.delete_segment(segment_id)
    if not success:
        raise HTTPException(status_code=404, detail="分段不存在")

    return ResponseModel(message="删除成功")


# ==================== 标注（Q&A）====================

@router.get("/{kb_id}/annotations", response_model=PaginatedResponse[AnnotationResponse], summary="标注列表")
async def list_annotations(
    kb_id: str,
    page: int = Query(1, ge=1),
    page_size: int = Query(20, ge=1, le=100, alias="pageSize"),
    keyword: Optional[str] = Query(None, description="搜索关键词"),
    db: AsyncSession = Depends(get_db),
):
    """获取知识库标注列表"""
    from sqlalchemy import or_, func as sa_func

    query = select(KnowledgeAnnotation).where(
        KnowledgeAnnotation.knowledge_base_id == kb_id,
        KnowledgeAnnotation.is_deleted == False,
    )
    if keyword:
        query = query.where(or_(
            KnowledgeAnnotation.question.contains(keyword),
            KnowledgeAnnotation.answer.contains(keyword),
        ))

    # 总数
    count_query = select(sa_func.count()).select_from(query.subquery())
    total = (await db.execute(count_query)).scalar() or 0

    # 分页
    query = query.order_by(KnowledgeAnnotation.sys_create_datetime.desc())
    query = query.offset((page - 1) * page_size).limit(page_size)
    result = await db.execute(query)
    items = result.scalars().all()

    return PaginatedResponse(items=items, total=total)


@router.post("/{kb_id}/annotations", response_model=AnnotationResponse, summary="创建标注")
async def create_annotation(
    kb_id: str,
    data: AnnotationCreateInput,
    background_tasks: BackgroundTasks,
    db: AsyncSession = Depends(get_db),
):
    """创建知识库标注（Q&A 对）"""
    annotation = KnowledgeAnnotation(
        knowledge_base_id=kb_id,
        question=data.question,
        answer=data.answer,
    )
    db.add(annotation)
    await db.commit()
    await db.refresh(annotation)

    # 后台向量化 question
    background_tasks.add_task(_vectorize_annotation, kb_id, str(annotation.id))

    return annotation


@router.put("/{kb_id}/annotations/{annotation_id}", response_model=AnnotationResponse, summary="更新标注")
async def update_annotation(
    kb_id: str,
    annotation_id: str,
    data: AnnotationUpdateInput,
    background_tasks: BackgroundTasks,
    db: AsyncSession = Depends(get_db),
):
    """更新标注"""
    result = await db.execute(
        select(KnowledgeAnnotation).where(
            KnowledgeAnnotation.id == annotation_id,
            KnowledgeAnnotation.knowledge_base_id == kb_id,
            KnowledgeAnnotation.is_deleted == False,
        )
    )
    annotation = result.scalar_one_or_none()
    if not annotation:
        raise HTTPException(status_code=404, detail="标注不存在")

    question_changed = False
    update_data = data.model_dump(exclude_unset=True)
    for key, value in update_data.items():
        if key == 'question' and value != annotation.question:
            question_changed = True
        setattr(annotation, key, value)

    await db.commit()
    await db.refresh(annotation)

    # 如果 question 变了，重新向量化
    if question_changed:
        background_tasks.add_task(_vectorize_annotation, kb_id, annotation_id)

    return annotation


@router.delete("/{kb_id}/annotations/{annotation_id}", response_model=ResponseModel, summary="删除标注")
async def delete_annotation(
    kb_id: str,
    annotation_id: str,
    db: AsyncSession = Depends(get_db),
):
    """删除标注"""
    result = await db.execute(
        select(KnowledgeAnnotation).where(
            KnowledgeAnnotation.id == annotation_id,
            KnowledgeAnnotation.knowledge_base_id == kb_id,
            KnowledgeAnnotation.is_deleted == False,
        )
    )
    annotation = result.scalar_one_or_none()
    if not annotation:
        raise HTTPException(status_code=404, detail="标注不存在")

    annotation.is_deleted = True
    await db.commit()

    # 从 Qdrant 删除向量
    try:
        from ai_platform.knowledge.vector_store import get_vector_store
        vector_store = get_vector_store()
        await vector_store.delete(kb_id, [annotation_id])
    except Exception as e:
        logger.warning(f"删除标注向量失败: {e}")

    return ResponseModel(message="删除成功")


@router.post("/{kb_id}/annotations/reindex", response_model=ResponseModel, summary="重新向量化所有标注")
async def reindex_annotations(
    kb_id: str,
    background_tasks: BackgroundTasks,
    db: AsyncSession = Depends(get_db),
):
    """重新向量化知识库中所有 pending/failed 状态的标注"""
    result = await db.execute(
        select(KnowledgeAnnotation).where(
            KnowledgeAnnotation.knowledge_base_id == kb_id,
            KnowledgeAnnotation.is_deleted == False,
            KnowledgeAnnotation.embedding_status.in_(['pending', 'failed']),
        )
    )
    annotations = result.scalars().all()
    for ann in annotations:
        background_tasks.add_task(_vectorize_annotation, kb_id, str(ann.id))

    return ResponseModel(message=f"已提交 {len(annotations)} 条标注的向量化任务")


async def _vectorize_annotation(kb_id: str, annotation_id: str):
    """后台任务：向量化标注的 question"""
    from app.database import AsyncSessionLocal
    from ai_platform.knowledge.services.embedding_service import EmbeddingService
    from ai_platform.knowledge.vector_store import get_vector_store, VectorPoint

    try:
        async with AsyncSessionLocal() as db:
            # 获取知识库配置
            kb_result = await db.execute(
                select(KnowledgeBase).where(KnowledgeBase.id == kb_id, KnowledgeBase.is_deleted == False)
            )
            kb = kb_result.scalar_one_or_none()
            if not kb or not kb.embedding_model_id:
                return

            # 获取标注
            ann_result = await db.execute(
                select(KnowledgeAnnotation).where(
                    KnowledgeAnnotation.id == annotation_id,
                    KnowledgeAnnotation.is_deleted == False,
                )
            )
            annotation = ann_result.scalar_one_or_none()
            if not annotation:
                return

            # 向量化 question
            embedding_service = EmbeddingService(db)
            embedding = await embedding_service.embed_text(
                model_id=kb.embedding_model_id,
                text=annotation.question,
                dimensions=kb.embedding_dimensions if kb.embedding_dimensions else None,
            )

            # 写入 Qdrant（使用 annotation_id 作为向量 ID）
            vector_store = get_vector_store()
            vector_size = kb.embedding_dimensions or 1536
            await vector_store.ensure_collection(kb_id, vector_size)
            await vector_store.upsert(kb_id, [VectorPoint(
                id=annotation_id,
                vector=embedding,
                payload={
                    'knowledge_base_id': kb_id,
                    'type': 'annotation',
                },
            )])

            annotation.embedding_status = 'completed'
            await db.commit()
            logger.info(f"标注 {annotation_id} 向量化完成")

    except Exception as e:
        logger.error(f"标注向量化失败: {e}")
        try:
            async with AsyncSessionLocal() as db:
                ann_result = await db.execute(
                    select(KnowledgeAnnotation).where(KnowledgeAnnotation.id == annotation_id)
                )
                annotation = ann_result.scalar_one_or_none()
                if annotation:
                    annotation.embedding_status = 'failed'
                    await db.commit()
        except Exception:
            pass


# ==================== 索引进度 SSE ====================

@router.get("/{kb_id}/indexing/progress", summary="订阅索引进度（SSE）")
async def indexing_progress_sse(kb_id: str):
    """通过 Server-Sent Events 推送索引进度"""
    from fastapi.responses import StreamingResponse
    from ai_platform.knowledge.services.indexing_progress_service import IndexingProgressService
    import json
    import asyncio

    async def event_generator():
        try:
            async for event in IndexingProgressService.subscribe(kb_id):
                data = json.dumps(event, ensure_ascii=False)
                yield f"data: {data}\n\n"
                # 如果收到 completed 或 failed，延迟后结束
                if event.get("step") in ("completed", "failed"):
                    await asyncio.sleep(0.5)
        except asyncio.CancelledError:
            pass
        except Exception as e:
            logger.warning(f"SSE 连接异常: {e}")

    return StreamingResponse(
        event_generator(),
        media_type="text/event-stream",
        headers={
            "Cache-Control": "no-cache",
            "Connection": "keep-alive",
            "X-Accel-Buffering": "no",
        },
    )


# ==================== 检索 ====================

@router.post("/retrieve", response_model=RetrievalResponse, summary="知识库检索")
async def retrieve(data: RetrievalInput, db: AsyncSession = Depends(get_db)):
    """检索知识库"""
    import time
    from ai_platform.knowledge.services.retrieval_service import RetrievalService

    start_time = time.time()
    service = RetrievalService(db)

    # 解析实际使用的检索模式和 rerank 配置
    actual_retrieval_mode = data.retrieval_mode or ''
    actual_rerank = data.rerank_enabled
    if not actual_retrieval_mode or actual_rerank is None:
        kb_result = await db.execute(
            select(KnowledgeBase).where(
                KnowledgeBase.id == data.knowledge_base_ids[0],
                KnowledgeBase.is_deleted == False
            )
        )
        first_kb = kb_result.scalar_one_or_none()
        if first_kb:
            if not actual_retrieval_mode:
                actual_retrieval_mode = first_kb.retrieval_mode or 'hybrid'
            if actual_rerank is None:
                actual_rerank = first_kb.rerank_enabled or False

    try:
        results = await service.retrieve(
            query=data.query,
            knowledge_base_ids=data.knowledge_base_ids,
            top_k=data.top_k,
            score_threshold=data.score_threshold,
            retrieval_mode=data.retrieval_mode,
            rerank_enabled=data.rerank_enabled,
            rerank_model_id=data.rerank_model_id,
            metadata_filter=data.metadata_filter,
        )
    except Exception as e:
        logger.exception(f'检索失败: {e}')
        raise HTTPException(status_code=500, detail=f"检索失败: {str(e)}")

    elapsed = int((time.time() - start_time) * 1000)

    # 记录检索日志
    try:
        from ai_platform.knowledge.models import KnowledgeRetrievalLog
        log = KnowledgeRetrievalLog(
            query=data.query,
            knowledge_base_ids=data.knowledge_base_ids,
            retrieval_mode=actual_retrieval_mode or 'hybrid',
            top_k=data.top_k,
            score_threshold=data.score_threshold,
            result_count=len(results),
            results=[
                {'segment_id': r.segment_id, 'score': r.score, 'kb_id': r.knowledge_base_id}
                for r in results
            ],
            rerank_applied='true' if actual_rerank else 'false',
            elapsed_time=elapsed,
            source='api',
        )
        db.add(log)
        await db.commit()
    except Exception as e:
        logger.warning(f'记录检索日志失败: {e}')

    return RetrievalResponse(
        results=results,
        total=len(results),
        query=data.query,
        elapsed_time=elapsed,
        retrieval_mode=actual_retrieval_mode or 'hybrid',
        rerank_applied=actual_rerank or False,
    )


# ==================== 命中统计 ====================

@router.get("/{kb_id}/hit-stats", summary="分段命中统计（热力图数据）")
async def get_hit_stats(
    kb_id: str,
    top_n: int = Query(50, ge=1, le=200, alias="topN", description="返回 Top N 命中分段"),
    include_zero: bool = Query(False, alias="includeZero", description="是否包含未命中分段"),
    db: AsyncSession = Depends(get_db),
):
    """
    获取知识库分段命中统计，用于热力图可视化。
    返回按命中次数降序排列的分段列表。
    """
    from ai_platform.knowledge.models import KnowledgeSegment, KnowledgeDocument
    from sqlalchemy import func as sa_func

    query = select(
        KnowledgeSegment.id,
        KnowledgeSegment.document_id,
        KnowledgeSegment.position,
        KnowledgeSegment.content,
        KnowledgeSegment.hit_count,
        KnowledgeSegment.enabled,
    ).where(
        KnowledgeSegment.knowledge_base_id == kb_id,
        KnowledgeSegment.is_deleted == False,
    )

    if not include_zero:
        query = query.where(KnowledgeSegment.hit_count > 0)

    query = query.order_by(KnowledgeSegment.hit_count.desc()).limit(top_n)
    result = await db.execute(query)
    rows = result.all()

    # 获取文档名称
    doc_ids = list({r.document_id for r in rows})
    doc_names = {}
    if doc_ids:
        doc_result = await db.execute(
            select(KnowledgeDocument.id, KnowledgeDocument.name).where(
                KnowledgeDocument.id.in_(doc_ids)
            )
        )
        doc_names = {row.id: row.name for row in doc_result}

    # 总命中次数（用于计算百分比）
    total_hits_result = await db.execute(
        select(sa_func.coalesce(sa_func.sum(KnowledgeSegment.hit_count), 0)).where(
            KnowledgeSegment.knowledge_base_id == kb_id,
            KnowledgeSegment.is_deleted == False,
        )
    )
    total_hits = total_hits_result.scalar() or 0

    items = []
    for r in rows:
        items.append({
            "segment_id": r.id,
            "document_id": r.document_id,
            "document_name": doc_names.get(r.document_id, ""),
            "position": r.position,
            "content_preview": r.content[:100] if r.content else "",
            "hit_count": r.hit_count or 0,
            "hit_percentage": round((r.hit_count or 0) / total_hits * 100, 2) if total_hits > 0 else 0,
            "enabled": r.enabled,
        })

    # 未命中分段统计
    zero_hit_result = await db.execute(
        select(sa_func.count()).where(
            KnowledgeSegment.knowledge_base_id == kb_id,
            KnowledgeSegment.is_deleted == False,
            KnowledgeSegment.hit_count == 0,
        )
    )
    zero_hit_count = zero_hit_result.scalar() or 0

    return {
        "items": items,
        "total_hits": total_hits,
        "total_segments": len(items) + zero_hit_count if not include_zero else len(items),
        "zero_hit_count": zero_hit_count,
    }


# ==================== 辅助函数 ====================

def _build_kb_response(kb: KnowledgeBase, embedding_model_name: str = "") -> dict:
    return {
        "id": kb.id,
        "application_id": kb.application_id,
        "is_global": kb.is_global or False,
        "name": kb.name,
        "code": kb.code,
        "description": kb.description or "",
        "icon": kb.icon or "",
        "embedding_model_id": kb.embedding_model_id,
        "embedding_model_name": embedding_model_name,
        "embedding_dimensions": kb.embedding_dimensions or 1536,
        "chunk_strategy": kb.chunk_strategy or "recursive",
        "chunk_size": kb.chunk_size or 500,
        "chunk_overlap": kb.chunk_overlap or 50,
        "separator": kb.separator,
        "retrieval_mode": kb.retrieval_mode or "hybrid",
        "top_k": kb.top_k or 5,
        "score_threshold": kb.score_threshold or 0.5,
        "rerank_enabled": kb.rerank_enabled or False,
        "rerank_model_id": kb.rerank_model_id,
        "retrieval_weight": kb.retrieval_weight or 1.0,
        "process_rules": kb.process_rules,
        "indexing_technique": kb.indexing_technique or "high_quality",
        "document_count": kb.document_count or 0,
        "segment_count": kb.segment_count or 0,
        "total_token_count": kb.total_token_count or 0,
        "total_char_count": kb.total_char_count or 0,
        "status": kb.status or "active",
        "sort": kb.sort or 0,
        "sys_create_datetime": kb.sys_create_datetime,
        "sys_update_datetime": kb.sys_update_datetime,
    }


def _build_doc_response(doc: KnowledgeDocument) -> dict:
    return {
        "id": doc.id,
        "knowledge_base_id": doc.knowledge_base_id,
        "file_id": doc.file_id,
        "name": doc.name,
        "file_type": doc.file_type or "",
        "file_size": doc.file_size or 0,
        "content_hash": doc.content_hash or "",
        "segment_count": doc.segment_count or 0,
        "token_count": doc.token_count or 0,
        "char_count": doc.char_count or 0,
        "status": doc.status or "pending",
        "error_message": doc.error_message or "",
        "enabled": doc.enabled if doc.enabled is not None else True,
        "indexing_started_at": doc.indexing_started_at,
        "indexing_completed_at": doc.indexing_completed_at,
        "sys_create_datetime": doc.sys_create_datetime,
        "sys_update_datetime": doc.sys_update_datetime,
    }


def _build_doc_list_response(doc: KnowledgeDocument) -> dict:
    return {
        "id": doc.id,
        "knowledge_base_id": doc.knowledge_base_id,
        "file_id": doc.file_id,
        "name": doc.name,
        "file_type": doc.file_type or "",
        "file_size": doc.file_size or 0,
        "segment_count": doc.segment_count or 0,
        "token_count": doc.token_count or 0,
        "status": doc.status or "pending",
        "duplicate_warning": doc.duplicate_warning or "",
        "enabled": doc.enabled if doc.enabled is not None else True,
        "sys_create_datetime": doc.sys_create_datetime,
    }


def _build_segment_list_response(seg, doc_name_map: dict) -> dict:
    return {
        "id": seg.id,
        "document_id": seg.document_id,
        "document_name": doc_name_map.get(seg.document_id, ""),
        "position": seg.position or 0,
        "content": seg.content or "",
        "answer": getattr(seg, 'answer', None) or "",
        "token_count": seg.token_count or 0,
        "char_count": seg.char_count or 0,
        "word_count": getattr(seg, 'word_count', None) or 0,
        "page_number": getattr(seg, 'page_number', None),
        "keywords": (seg.keywords or []) if hasattr(seg, 'keywords') else [],
        "extra_metadata": getattr(seg, 'extra_metadata', None),
        "enabled": seg.enabled if seg.enabled is not None else True,
        "hit_count": seg.hit_count or 0,
        "embedding_status": seg.embedding_status or "pending",
        "sys_create_datetime": seg.sys_create_datetime,
    }


async def _index_document_task(knowledge_base_id: str, document_id: str):
    """后台索引文档任务"""
    from app.database import AsyncSessionLocal
    from ai_platform.knowledge.services.indexing_service import IndexingService

    async with AsyncSessionLocal() as db:
        try:
            service = IndexingService(db)
            await service.index_document(knowledge_base_id, document_id)
        except Exception as e:
            logger.exception(f'后台索引文档失败: {e}')


async def _batch_index_documents_task(knowledge_base_id: str, document_ids: list, max_concurrency: int = 3):
    """后台并发索引多个文档"""
    import asyncio
    from app.database import AsyncSessionLocal
    from ai_platform.knowledge.services.indexing_service import IndexingService

    semaphore = asyncio.Semaphore(max_concurrency)

    async def _index_one(doc_id: str):
        async with semaphore:
            async with AsyncSessionLocal() as db:
                try:
                    service = IndexingService(db)
                    await service.index_document(knowledge_base_id, doc_id)
                except Exception as e:
                    logger.exception(f'并发索引文档失败 (doc_id={doc_id}): {e}')

    await asyncio.gather(*[_index_one(doc_id) for doc_id in document_ids])