feat: 增加 Agent Excel 批量测试

ai_knowledge_base_v2/apps/backend/app/services/admin_permission_service.py

@@ -12,7 +12,7 @@ PERMISSION_TREE = [
     {"code": "users", "name": "用户管理", "children": [{"code": "users.view", "name": "查看用户"}, {"code": "users.create", "name": "新增/导入"}, {"code": "users.edit", "name": "编辑/权益续期"}, {"code": "users.delete", "name": "删除用户"}]},
     {"code": "entitlements", "name": "权益管理", "children": [{"code": "entitlements.view", "name": "查看权益"}, {"code": "entitlements.edit", "name": "编辑权益"}]},
     {"code": "knowledge", "name": "知识库管理", "children": [{"code": "knowledge.view", "name": "查看知识库"}, {"code": "knowledge.edit", "name": "新增/编辑/同步"}, {"code": "knowledge.publish", "name": "开放/归档"}, {"code": "knowledge.delete", "name": "删除知识库"}]},
-    {"code": "prompt", "name": "Agent 管理", "children": [{"code": "prompt.view", "name": "查看 Agent"}, {"code": "prompt.edit", "name": "编辑/测试 Agent"}]},
+    {"code": "prompt", "name": "Agent 管理", "children": [{"code": "prompt.view", "name": "查看 Agent"}, {"code": "prompt.edit", "name": "编辑/单条测试 Agent"}, {"code": "prompt.batch", "name": "批量测试/导出"}]},
     {"code": "models", "name": "模型管理", "children": [{"code": "models.view", "name": "查看模型"}, {"code": "models.edit", "name": "新增/编辑/测试"}, {"code": "models.delete", "name": "删除模型"}]},
     {"code": "content-generation", "name": "内容生成", "children": [{"code": "content-generation.view", "name": "查看配置"}, {"code": "content-generation.edit", "name": "编辑/测试配置"}]},
     {"code": "configs", "name": "系统配置", "children": [{"code": "configs.view", "name": "查看配置"}, {"code": "configs.edit", "name": "修改配置"}]},

ai_knowledge_base_v2/apps/backend/app/services/agent_batch_test_service.py

@@ -0,0 +1,274 @@
+from __future__ import annotations
+
+import json
+from datetime import UTC, datetime
+from io import BytesIO
+
+from fastapi import HTTPException, UploadFile, status
+from openpyxl import Workbook, load_workbook
+from openpyxl.styles import Alignment, Font, PatternFill
+from openpyxl.worksheet.table import Table, TableStyleInfo
+from sqlalchemy import func, select
+from sqlalchemy.orm import Session
+
+from app.models.admin import Admin
+from app.models.agent_batch import AgentBatchTest, AgentBatchTestItem
+from app.models.ai_config import ModelConfig
+from app.models.knowledge import Knowledge
+
+
+MAX_BATCH_QUESTIONS = 200
+MAX_QUESTION_LENGTH = 2000
+TEMPLATE_HEADERS = ["序号", "问题"]
+
+
+class AgentBatchTestService:
+    @staticmethod
+    def parse_questions(file: UploadFile) -> list[dict]:
+        filename = file.filename or ""
+        if not filename.lower().endswith(".xlsx"):
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="请上传 .xlsx 格式的 Excel 文件")
+        content = file.file.read()
+        if not content:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Excel 文件为空")
+        if len(content) > 5 * 1024 * 1024:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Excel 文件不能超过 5MB")
+        try:
+            workbook = load_workbook(BytesIO(content), read_only=True, data_only=False)
+        except Exception as exc:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Excel 文件解析失败，请重新下载模板填写") from exc
+        sheet = workbook.active
+        if sheet.max_row > 1000:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Excel 行数异常，请删除多余空行后重试")
+        headers = [str(value or "").strip() for value in next(sheet.iter_rows(min_row=1, max_row=1, values_only=True), ())]
+        header_map = {name: index for index, name in enumerate(headers) if name}
+        if "问题" not in header_map:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="模板缺少必填列：问题")
+        rows: list[dict] = []
+        seen: set[str] = set()
+        for row_number, values in enumerate(sheet.iter_rows(min_row=2, max_row=min(sheet.max_row, 1000), values_only=True), start=2):
+            raw_question = _value_at(values, header_map["问题"])
+            if isinstance(raw_question, str) and raw_question.startswith("="):
+                raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=f"第 {row_number} 行问题不能使用 Excel 公式")
+            question = _cell_text(raw_question)
+            external_no = _cell_text(_value_at(values, header_map.get("序号"))) or str(row_number - 1)
+            if not question:
+                continue
+            if len(question) > MAX_QUESTION_LENGTH:
+                raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=f"第 {row_number} 行问题超过 {MAX_QUESTION_LENGTH} 字")
+            normalized = " ".join(question.split())
+            if normalized in seen:
+                raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=f"第 {row_number} 行问题与前面重复")
+            seen.add(normalized)
+            rows.append({"rowNumber": row_number, "externalNo": external_no[:100], "question": question})
+            if len(rows) > MAX_BATCH_QUESTIONS:
+                raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=f"单次最多导入 {MAX_BATCH_QUESTIONS} 个问题")
+        if not rows:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Excel 中没有可测试的问题")
+        return rows
+
+    @staticmethod
+    def create_job(
+        db: Session,
+        *,
+        admin: Admin,
+        filename: str,
+        rows: list[dict],
+        config: dict,
+    ) -> AgentBatchTest:
+        prompt_content = str(config.get("promptContent") or "").strip()
+        if not prompt_content:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="主提示词不能为空")
+        try:
+            model_id = int(config.get("modelId"))
+        except (TypeError, ValueError) as exc:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="请选择批量测试模型") from exc
+        model = db.get(ModelConfig, model_id)
+        if model is None:
+            raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="批量测试模型不存在")
+        knowledge_ids = _positive_int_list(config.get("knowledgeIds"))
+        if knowledge_ids:
+            found = list(db.scalars(select(Knowledge).where(Knowledge.id.in_(knowledge_ids))))
+            if len(found) != len(knowledge_ids):
+                raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="所选知识库中包含已删除的数据")
+            knowledge_names = [item.name for item in sorted(found, key=lambda item: knowledge_ids.index(item.id))]
+        else:
+            knowledge_names = ["全部正式开放知识库"]
+        generation_config = {
+            "temperature": config.get("temperature"),
+            "topP": config.get("topP"),
+            "topK": config.get("topK"),
+            "presencePenalty": config.get("presencePenalty"),
+            "frequencyPenalty": config.get("frequencyPenalty"),
+            "maxToken": config.get("maxToken") or 8192,
+            "streamEnabled": int(config.get("streamEnabled", 1)),
+            "reasoningVisible": 0,
+            "responseDepth": int(config.get("responseDepth", 35)),
+        }
+        job = AgentBatchTest(
+            name=str(config.get("name") or filename.rsplit(".", 1)[0] or "Agent 批量测试")[:160],
+            original_filename=filename[:255],
+            total_count=len(rows),
+            prompt_content=prompt_content,
+            model_id=model.id,
+            model_name=model.display_name or model.model_name,
+            knowledge_ids=json.dumps(knowledge_ids, ensure_ascii=False),
+            knowledge_names=json.dumps(knowledge_names, ensure_ascii=False),
+            generation_config=json.dumps(generation_config, ensure_ascii=False),
+            created_by=admin.id,
+        )
+        db.add(job)
+        db.flush()
+        db.add_all([
+            AgentBatchTestItem(
+                batch_id=job.id,
+                row_number=row["rowNumber"],
+                external_no=row["externalNo"],
+                question=row["question"],
+            )
+            for row in rows
+        ])
+        db.flush()
+        return job
+
+    @staticmethod
+    def template_workbook() -> BytesIO:
+        workbook = Workbook()
+        sheet = workbook.active
+        sheet.title = "批量测试问题"
+        sheet.append(TEMPLATE_HEADERS)
+        sheet.append([1, "请填写需要测试的问题，保留一行一个问题。"])
+        sheet.append([2, "例如：遇到情绪焦虑时，可以先做什么？"])
+        _style_sheet(sheet, widths=(12, 72), table_ref="A1:B3", table_name="AgentBatchQuestionTable")
+        note = workbook.create_sheet("填写说明")
+        note.append(["填写规则", "说明"])
+        note.append(["必填列", "问题"])
+        note.append(["单次数量", f"最多 {MAX_BATCH_QUESTIONS} 个问题"])
+        note.append(["问题长度", f"每个问题最多 {MAX_QUESTION_LENGTH} 字"])
+        note.append(["重复问题", "同一文件内不允许重复"])
+        note.append(["不要修改", "不要修改工作表名称和表头；可以删除示例行后填写"])
+        _style_sheet(note, widths=(18, 72), table_ref="A1:B6", table_name="AgentBatchInstructions")
+        stream = BytesIO()
+        workbook.save(stream)
+        stream.seek(0)
+        return stream
+
+    @staticmethod
+    def export_workbook(job: AgentBatchTest, items: list[AgentBatchTestItem]) -> BytesIO:
+        workbook = Workbook()
+        sheet = workbook.active
+        sheet.title = "批量测试结果"
+        headers = ["序号", "问题", "答案", "状态", "失败原因", "实际模型", "召回数量", "耗时(秒)"]
+        sheet.append(headers)
+        status_labels = {"success": "成功", "failed": "失败", "cancelled": "已取消", "pending": "等待中", "running": "生成中"}
+        for item in items:
+            sheet.append([
+                item.external_no or item.row_number - 1,
+                _excel_safe(item.question),
+                _excel_safe(item.answer or ""),
+                status_labels.get(item.status, item.status),
+                _excel_safe(item.error_message or ""),
+                _excel_safe(item.model_name or job.model_name),
+                item.retrieve_count,
+                round(item.duration_ms / 1000, 2) if item.duration_ms is not None else None,
+            ])
+        end_row = max(2, len(items) + 1)
+        _style_sheet(sheet, widths=(12, 42, 80, 12, 36, 22, 12, 12), table_ref=f"A1:H{end_row}", table_name="AgentBatchResultTable")
+        for row in sheet.iter_rows(min_row=2, max_row=end_row):
+            row[1].alignment = Alignment(vertical="top", wrap_text=True)
+            row[2].alignment = Alignment(vertical="top", wrap_text=True)
+            row[4].alignment = Alignment(vertical="top", wrap_text=True)
+        summary = workbook.create_sheet("任务信息")
+        summary.append(["项目", "内容"])
+        summary.append(["任务名称", job.name])
+        summary.append(["任务状态", job.status])
+        summary.append(["问题总数", job.total_count])
+        summary.append(["成功", job.success_count])
+        summary.append(["失败", job.failed_count])
+        summary.append(["测试模型", job.model_name])
+        summary.append(["知识库", "、".join(json.loads(job.knowledge_names or "[]"))])
+        summary.append(["创建时间", job.created_at])
+        summary.append(["完成时间", job.finished_at])
+        _style_sheet(summary, widths=(20, 86), table_ref="A1:B10", table_name="AgentBatchSummary")
+        summary.column_dimensions["B"].width = 86
+        stream = BytesIO()
+        workbook.save(stream)
+        stream.seek(0)
+        return stream
+
+
+def job_dict(job: AgentBatchTest) -> dict:
+    return {
+        "id": job.id,
+        "name": job.name,
+        "originalFilename": job.original_filename,
+        "status": job.status,
+        "totalCount": job.total_count,
+        "processedCount": job.processed_count,
+        "successCount": job.success_count,
+        "failedCount": job.failed_count,
+        "progress": round(job.processed_count * 100 / job.total_count) if job.total_count else 0,
+        "modelId": job.model_id,
+        "modelName": job.model_name,
+        "knowledgeIds": json.loads(job.knowledge_ids or "[]"),
+        "knowledgeNames": json.loads(job.knowledge_names or "[]"),
+        "createdAt": job.created_at,
+        "startedAt": job.started_at,
+        "finishedAt": job.finished_at,
+    }
+
+
+def item_dict(item: AgentBatchTestItem) -> dict:
+    return {
+        "id": item.id,
+        "rowNumber": item.row_number,
+        "externalNo": item.external_no,
+        "question": item.question,
+        "answer": item.answer,
+        "status": item.status,
+        "errorMessage": item.error_message,
+        "modelName": item.model_name,
+        "retrieveCount": item.retrieve_count,
+        "durationMs": item.duration_ms,
+    }
+
+
+def _style_sheet(sheet, *, widths: tuple[int, ...], table_ref: str, table_name: str) -> None:
+    sheet.freeze_panes = "A2"
+    sheet.sheet_view.showGridLines = False
+    header_fill = PatternFill("solid", fgColor="176B57")
+    for cell in sheet[1]:
+        cell.fill = header_fill
+        cell.font = Font(color="FFFFFF", bold=True)
+        cell.alignment = Alignment(vertical="center")
+    sheet.row_dimensions[1].height = 28
+    for index, width in enumerate(widths, start=1):
+        sheet.column_dimensions[chr(64 + index)].width = width
+    table = Table(displayName=table_name, ref=table_ref)
+    table.tableStyleInfo = TableStyleInfo(name="TableStyleMedium4", showRowStripes=True, showFirstColumn=False, showLastColumn=False)
+    sheet.add_table(table)
+
+
+def _cell_text(value) -> str:
+    if value is None:
+        return ""
+    return str(value).strip()
+
+
+def _value_at(row: tuple, index: int | None):
+    return row[index] if index is not None and index < len(row) else None
+
+
+def _positive_int_list(value) -> list[int]:
+    if not isinstance(value, list):
+        return []
+    try:
+        result = list(dict.fromkeys(int(item) for item in value if int(item) > 0))
+    except (TypeError, ValueError) as exc:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="知识库参数格式不正确") from exc
+    return result
+
+
+def _excel_safe(value: str) -> str:
+    text = str(value or "")
+    return f"'{text}" if text.lstrip().startswith(("=", "+", "-", "@")) else text

ai_knowledge_base_v2/apps/backend/app/services/agent_batch_test_worker.py

@@ -0,0 +1,227 @@
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+import os
+import socket
+import time
+import uuid
+from datetime import UTC, datetime, timedelta
+
+from sqlalchemy import func, or_, select
+from sqlalchemy.orm import Session
+
+from app.core.config import get_settings
+from app.core.database import SessionLocal
+from app.models.admin import Admin
+from app.models.agent_batch import AgentBatchTest, AgentBatchTestItem
+from app.schemas.admin import AgentDebugRequest
+from app.services.agent_debug_service import AgentDebugService
+
+
+logger = logging.getLogger(__name__)
+
+
+class AgentBatchTestWorker:
+    """Persistent, restart-safe queue for Agent batch questions."""
+
+    @classmethod
+    async def run_forever(cls) -> None:
+        settings = get_settings()
+        if not settings.agent_batch_worker_enabled:
+            logger.info("agent batch test worker disabled")
+            return
+        worker_id = f"{socket.gethostname()}:{os.getpid()}:{uuid.uuid4().hex[:8]}"
+        while True:
+            try:
+                processed = await cls.run_once(worker_id)
+            except Exception:
+                processed = False
+                logger.exception("agent batch test worker iteration failed")
+            await asyncio.sleep(0 if processed else max(1, settings.agent_batch_poll_seconds))
+
+    @classmethod
+    async def run_once(cls, worker_id: str) -> bool:
+        now = _now()
+        with SessionLocal() as db:
+            recovered_batch_ids = cls.recover_stale_items(db, now=now)
+            db.commit()
+            for batch_id in recovered_batch_ids:
+                cls.refresh_job(db, batch_id)
+        with SessionLocal() as db:
+            item_id = cls.claim_next(db, worker_id=worker_id, now=now)
+        if item_id is None:
+            return False
+        await cls.execute_claimed(item_id=item_id, worker_id=worker_id)
+        return True
+
+    @staticmethod
+    def claim_next(db: Session, *, worker_id: str, now: datetime | None = None) -> int | None:
+        current = now or _now()
+        item = db.scalar(
+            select(AgentBatchTestItem)
+            .join(AgentBatchTest, AgentBatchTest.id == AgentBatchTestItem.batch_id)
+            .where(
+                AgentBatchTest.status.in_(("pending", "running")),
+                AgentBatchTestItem.status == "pending",
+                AgentBatchTestItem.attempt_count < AgentBatchTestItem.max_attempts,
+                or_(AgentBatchTestItem.next_run_at.is_(None), AgentBatchTestItem.next_run_at <= current),
+            )
+            .order_by(AgentBatchTestItem.batch_id.asc(), AgentBatchTestItem.row_number.asc())
+            .with_for_update(skip_locked=True)
+            .limit(1)
+        )
+        if item is None:
+            db.rollback()
+            return None
+        job = db.get(AgentBatchTest, item.batch_id)
+        item.status = "running"
+        item.attempt_count += 1
+        item.locked_at = current
+        item.locked_by = worker_id
+        item.started_at = current
+        if job is not None and job.status == "pending":
+            job.status = "running"
+            job.started_at = current
+            db.add(job)
+        db.add(item)
+        db.commit()
+        return item.id
+
+    @classmethod
+    async def execute_claimed(cls, *, item_id: int, worker_id: str) -> None:
+        with SessionLocal() as db:
+            item = db.get(AgentBatchTestItem, item_id)
+            if item is None or item.status != "running" or item.locked_by != worker_id:
+                return
+            job = db.get(AgentBatchTest, item.batch_id)
+            admin = db.get(Admin, job.created_by) if job is not None else None
+            if job is None or admin is None:
+                cls._finish_item(db, item, error="批量任务或创建管理员不存在")
+                return
+            if job.status == "cancelled":
+                item.status = "cancelled"
+                item.finished_at = _now()
+                item.locked_at = None
+                item.locked_by = None
+                db.add(item)
+                db.commit()
+                return
+            payload = _payload_from(job, item)
+            started = time.monotonic()
+            try:
+                result = await AgentDebugService.run_complete(payload, db, admin)
+                duration_ms = round((time.monotonic() - started) * 1000)
+                item.answer = result["answer"]
+                item.model_name = result.get("modelName") or job.model_name
+                item.knowledge_ids = json.dumps(result.get("knowledgeIds") or [], ensure_ascii=False)
+                item.retrieve_count = result.get("retrieveCount")
+                item.duration_ms = duration_ms
+                item.status = "success"
+                item.error_message = None
+                item.finished_at = _now()
+                item.next_run_at = None
+                item.locked_at = None
+                item.locked_by = None
+                db.add(item)
+                db.commit()
+            except Exception as exc:
+                db.rollback()
+                item = db.get(AgentBatchTestItem, item_id)
+                if item is None:
+                    return
+                item.duration_ms = round((time.monotonic() - started) * 1000)
+                if item.attempt_count < item.max_attempts:
+                    item.status = "pending"
+                    item.next_run_at = _now() + timedelta(seconds=15 * item.attempt_count)
+                    item.finished_at = None
+                else:
+                    item.status = "failed"
+                    item.next_run_at = None
+                    item.finished_at = _now()
+                item.error_message = (str(exc) or "Agent 生成失败")[:2000]
+                item.locked_at = None
+                item.locked_by = None
+                db.add(item)
+                db.commit()
+            cls.refresh_job(db, job.id)
+
+    @staticmethod
+    def refresh_job(db: Session, batch_id: int) -> None:
+        job = db.get(AgentBatchTest, batch_id)
+        if job is None or job.status == "cancelled":
+            return
+        counts = dict(
+            db.execute(
+                select(AgentBatchTestItem.status, func.count(AgentBatchTestItem.id))
+                .where(AgentBatchTestItem.batch_id == batch_id)
+                .group_by(AgentBatchTestItem.status)
+            ).all()
+        )
+        job.success_count = counts.get("success", 0)
+        job.failed_count = counts.get("failed", 0)
+        job.processed_count = job.success_count + job.failed_count
+        if job.processed_count >= job.total_count:
+            job.status = "completed" if job.failed_count == 0 else "completed_with_errors"
+            job.finished_at = _now()
+        else:
+            job.status = "running"
+        db.add(job)
+        db.commit()
+
+    @staticmethod
+    def recover_stale_items(db: Session, *, now: datetime | None = None) -> set[int]:
+        current = now or _now()
+        stale_before = current - timedelta(minutes=max(5, get_settings().agent_batch_stale_minutes))
+        items = list(
+            db.scalars(
+                select(AgentBatchTestItem)
+                .where(
+                    AgentBatchTestItem.status == "running",
+                    AgentBatchTestItem.locked_at.is_not(None),
+                    AgentBatchTestItem.locked_at < stale_before,
+                )
+                .order_by(AgentBatchTestItem.id.asc())
+                .limit(100)
+                .with_for_update(skip_locked=True)
+            )
+        )
+        affected_batch_ids: set[int] = set()
+        for item in items:
+            affected_batch_ids.add(item.batch_id)
+            item.status = "pending" if item.attempt_count < item.max_attempts else "failed"
+            item.error_message = "任务执行进程中断，系统已自动恢复" if item.status == "pending" else "任务执行进程连续中断"
+            item.next_run_at = current if item.status == "pending" else None
+            item.finished_at = current if item.status == "failed" else None
+            item.locked_at = None
+            item.locked_by = None
+            db.add(item)
+        return affected_batch_ids
+
+    @classmethod
+    def _finish_item(cls, db: Session, item: AgentBatchTestItem, *, error: str) -> None:
+        item.status = "failed"
+        item.error_message = error
+        item.finished_at = _now()
+        item.locked_at = None
+        item.locked_by = None
+        db.add(item)
+        db.commit()
+        cls.refresh_job(db, item.batch_id)
+
+
+def _payload_from(job: AgentBatchTest, item: AgentBatchTestItem) -> AgentDebugRequest:
+    generation = json.loads(job.generation_config or "{}")
+    return AgentDebugRequest(
+        promptContent=job.prompt_content,
+        modelId=job.model_id,
+        knowledgeIds=json.loads(job.knowledge_ids or "[]"),
+        question=item.question,
+        history=[],
+        **generation,
+    )
+
+
+def _now() -> datetime:
+    return datetime.now(UTC).replace(tzinfo=None)

ai_knowledge_base_v2/apps/backend/app/services/agent_debug_service.py

@@ -283,3 +283,20 @@ class AgentDebugService:
         except Exception as exc:
             db.rollback()
             yield {"type": "error", "message": str(exc) or "Agent 调试失败"}
+
+    @classmethod
+    async def run_complete(cls, payload: AgentDebugRequest, db: Session, current_admin: Admin) -> dict:
+        """Consume the same routed debug stream used by the UI and return one complete answer."""
+        answer_parts: list[str] = []
+        completion: dict = {}
+        async for event in cls.stream(payload, db, current_admin):
+            if event.get("type") == "content":
+                answer_parts.append(str(event.get("content") or ""))
+            elif event.get("type") == "complete":
+                completion = event
+            elif event.get("type") == "error":
+                raise RuntimeError(str(event.get("message") or "Agent 生成失败"))
+        answer = "".join(answer_parts).strip()
+        if not answer:
+            raise RuntimeError("模型未返回正式回答")
+        return {"answer": answer, **completion}