Agent/examples/demand/web_api.py

"""
demand Web API（异步任务：发起 -> 立即返回 task_id -> 另一个接口查询状态）
"""

import asyncio
import os
import importlib
import sys
from datetime import datetime, timedelta
from pathlib import Path
from typing import Any, Literal, Optional
from zoneinfo import ZoneInfo

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel

from examples.demand.db_manager import exist_cluster_tree

# 添加项目根目录到 Python 路径（与 run.py 保持一致）
sys.path.insert(0, str(Path(__file__).parent.parent.parent))

from examples.demand.zengzhang_prepare import zengzhang_prepare
from examples.demand.changwen_prepare import changwen_prepare
from examples.demand.mysql import mysql_db
from examples.demand.piaoquan_prepare import piaoquan_prepare
from examples.demand.run import _create_demand_task, main as run_demand
from examples.demand.zaozhongwanhao_extra import (
    CLUSTER_NAME as ZAOZHONGWANHAO_CLUSTER,
    maybe_finish_zaoan_from_cluster,
    maybe_run_zaozhongwanhao_extra,
    schedule_has_zaozhongwanhao,
)

app = FastAPI(title="demand web api")

GLOBAL_TREE_SCHEDULE_ITEM = {
    "cluster_name": "全局树",
    "platform_type": "piaoquan",
    "count": 50,
}

# APScheduler：使用动态导入避免环境未安装时直接导入失败
try:
    _aps_asyncio_mod = importlib.import_module("apscheduler.schedulers.asyncio")
    _aps_cron_mod = importlib.import_module("apscheduler.triggers.cron")
    AsyncIOScheduler = getattr(_aps_asyncio_mod, "AsyncIOScheduler")
    CronTrigger = getattr(_aps_cron_mod, "CronTrigger")
except Exception:  # pragma: no cover
    AsyncIOScheduler = None  # type: ignore[assignment]
    CronTrigger = None  # type: ignore[assignment]


class DemandStartRequest(BaseModel):
    cluster_name: str
    platform_type: Literal["piaoquan", "changwen"]
    count: int


def _get_demand_schedule_cluster_platform_list() -> list[dict]:
    """动态获取定时任务列表。"""
    schedule_items: list[dict] = []
    try:
        # 延迟导入：避免服务启动时因 ODPS 依赖缺失直接失败
        from examples.demand.data_query_tools import get_demand_merge_level2_names
        schedule_items = get_demand_merge_level2_names() or []
    except Exception as e:  # pragma: no cover
        print(f"获取定时任务列表失败: {e}")

    # 固定补充：全局树每天都要产出 50 条需求（以固定值覆盖动态值）
    schedule_items = [
        item for item in schedule_items
        if not (
            item.get("cluster_name") == GLOBAL_TREE_SCHEDULE_ITEM["cluster_name"]
            and item.get("platform_type") == GLOBAL_TREE_SCHEDULE_ITEM["platform_type"]
        )
    ]
    schedule_items.append(GLOBAL_TREE_SCHEDULE_ITEM.copy())
    return schedule_items

# 是否开启定时任务（可选，通过环境变量覆盖）
DEMAND_SCHEDULER_ENABLED: bool = os.getenv("DEMAND_SCHEDULER_ENABLED", "1").strip() == "1"

DEMAND_SCHEDULER_START_HOUR: int = 2

# 定时任务统一使用北京时间，避免服务器时区（如 UTC）带来的偏差
BEIJING_TZ = ZoneInfo("Asia/Shanghai")


def _get_today_time_window(now: datetime) -> tuple[datetime, datetime]:
    """返回今天的 [start, end) 时间窗口（本地时区）。"""
    start_of_today = datetime(year=now.year, month=now.month, day=now.day, tzinfo=now.tzinfo)
    end_of_today = start_of_today + timedelta(days=1)
    return start_of_today, end_of_today


async def demand_start_sync(cluster_name: str, platform_type: Literal["piaoquan", "changwen"], count) -> dict:
    """
    与 /demand/start 同一执行链路，但不创建后台任务：prepare -> create demand_task -> 串行 await run_demand。
    """
    # prepare 阶段是同步的（当前示例代码为 sync），这里保持同步串行语义
    execution_id = None
    task_id: Optional[int] = None
    try:
        if platform_type == "piaoquan":
            exist = exist_cluster_tree(cluster_name)
            if not exist:
                raise ValueError("获取聚类树失败")
            execution_id = piaoquan_prepare(cluster_name)
        elif platform_type == "changwen":
            execution_id = changwen_prepare(cluster_name)
        elif platform_type == "zengzhang":
            execution_id = zengzhang_prepare(cluster_name)

        if not execution_id:
            raise ValueError("获取 execution_id 失败")

        task_name = cluster_name[:32] if cluster_name else None
        task_id = _create_demand_task(
            execution_id=execution_id,
            name=task_name,
            platform=platform_type,
        )
        if not task_id:
            raise ValueError("创建 demand_task 失败")

        # run_once 内部 finally 会把 task 状态写回 MySQL
        result = await run_demand(
            cluster_name,
            platform_type,
            count,
            execution_id=execution_id,
            task_id=task_id,
        )
        return {"ok": True, "message": "调用成功", "task_id": task_id, "execution_id": execution_id, "result": result}
    except Exception as e:
        return {
            "ok": False,
            "message": f"执行失败: {e}",
            "task_id": task_id,
            "execution_id": execution_id,
            "result": None,
        }


def _today_has_status_0_or_1(cluster_name: str, platform_type: str, now: datetime) -> bool:
    """
    查找 demand_task：
    - 限制为今天（create_time）
    - name 与 platform 精确匹配
    - 若存在 status 为 0 或 1 的记录，则跳过
    """
    start_of_today, end_of_today = _get_today_time_window(now)
    # MySQL DATETIME 一般按无时区存储，这里使用北京时间对应的“本地时间”窗口做过滤
    start_of_today_naive = start_of_today.replace(tzinfo=None)
    end_of_today_naive = end_of_today.replace(tzinfo=None)
    return mysql_db.exists(
        "demand_task",
        where=(
            "name = %s "
            "AND platform = %s "
            "AND status IN (0, 1) "
            "AND create_time >= %s "
            "AND create_time < %s"
        ),
        where_params=(
            str(cluster_name)[:32],
            str(platform_type)[:32],
            start_of_today_naive,
            end_of_today_naive,
        ),
    )


async def demand_scheduled_run_once() -> None:
    """
    任务批处理（串行）：
    遍历配置列表 -> 查当天 demand_task -> 匹配 cluster_name/name & platform_type/platform
    若存在 status=0 或 1 的记录则跳过；否则执行一次 demand_start_sync。
    「早中晚好」：有品类产出则按权重取 top10；没有则只生成 10 条。再筛早安策略入库。
    """
    # ODPS 查询是阻塞调用，放到线程里避免阻塞事件循环
    demand_schedule_cluster_platform_list = await asyncio.to_thread(_get_demand_schedule_cluster_platform_list)
    demand_schedule_cluster_platform_list = demand_schedule_cluster_platform_list or []

    now = datetime.now(BEIJING_TZ)
    has_zao = schedule_has_zaozhongwanhao(demand_schedule_cluster_platform_list)
    await maybe_run_zaozhongwanhao_extra(demand_schedule_cluster_platform_list, now=now)

    zao_execution_id = None
    for item in demand_schedule_cluster_platform_list:
        cluster_name = item.get("cluster_name")
        platform_type = item.get("platform_type")
        count = item.get("count")
        if not cluster_name or platform_type not in ("piaoquan", "changwen"):
            continue

        if _today_has_status_0_or_1(cluster_name, platform_type, now=now):
            print(f"[scheduler] skip: cluster={cluster_name}, platform={platform_type} (today has status 0/1)")
            continue

        print(f"[scheduler] run: cluster={cluster_name}, platform={platform_type}, count={count}")
        result = await demand_start_sync(cluster_name=cluster_name, platform_type=platform_type, count=count)  # 串行执行
        if str(cluster_name).strip() == ZAOZHONGWANHAO_CLUSTER and isinstance(result, dict):
            zao_execution_id = result.get("execution_id") or zao_execution_id

    if has_zao:
        await maybe_finish_zaoan_from_cluster(execution_id=zao_execution_id, now=now)


_demand_scheduler: Optional[Any] = None
_demand_scheduler_lock = asyncio.Lock()


async def _demand_scheduler_job() -> None:
    """
    定时任务 job：
    - 串行执行（防止并发）
    - 遍历配置 -> 今日过滤 demand_task -> 跳过/执行
    """
    if _demand_scheduler_lock.locked():
        return
    # 指定日期跳过：北京时间 2026-04-08 不执行定时任务
    if datetime.now(BEIJING_TZ).date() == datetime(2026, 4, 8).date():
        print("[scheduler] skip: 2026-04-08")
        return
    async with _demand_scheduler_lock:
        await demand_scheduled_run_once()


@app.post("/demand/start")
async def demand_start(req: DemandStartRequest):
    # 注意：这里会同步计算 execution_id（prepare 阶段），随后 run_once 放到后台异步执行。
    if req.platform_type == "piaoquan":
        execution_id = piaoquan_prepare(req.cluster_name)
    else:
        execution_id = changwen_prepare(req.cluster_name)

    if not execution_id:
        raise HTTPException(status_code=400, detail="获取 execution_id 失败")

    task_name = req.cluster_name[:32] if req.cluster_name else None
    task_id = _create_demand_task(
        execution_id=execution_id,
        name=task_name,
        platform=req.platform_type,
    )
    if not task_id:
        raise HTTPException(status_code=500, detail="创建 demand_task 失败")

    async def _job():
        # run_once 内部会在 finally 里把 task 状态写回 MySQL。
        await run_demand(
            req.cluster_name,
            req.platform_type,
            req.count,
            execution_id=execution_id,
            task_id=task_id,
        )

    asyncio.create_task(_job())
    return {"ok": True, "message": "调用成功", "task_id": task_id, "execution_id": execution_id}


@app.on_event("startup")
async def _start_demand_scheduler() -> None:
    """启动定时任务（北京时间：2:00–22:30 每 30 分钟，不含 23 点与 0 点）。"""
    global _demand_scheduler
    if not DEMAND_SCHEDULER_ENABLED:
        return
    if _demand_scheduler is not None:
        return

    if AsyncIOScheduler is None or CronTrigger is None:
        # 依赖未安装则跳过定时任务
        print("[scheduler] apscheduler 未安装，跳过定时任务启动")
        return

    scheduler = AsyncIOScheduler(timezone=BEIJING_TZ)
    start_h = DEMAND_SCHEDULER_START_HOUR
    # DEMAND_SCHEDULER_START_HOUR–22：每 30 分钟（默认 2:00–22:30，不含 23 点）
    if start_h > 22:
        print(f"[scheduler] DEMAND_SCHEDULER_START_HOUR={start_h} > 22，无法注册 cron，跳过定时任务")
    else:
        scheduler.add_job(
            func=_demand_scheduler_job,
            trigger=CronTrigger(hour=f"{start_h}-22", minute="0,30", timezone=BEIJING_TZ),
            id="demand_scheduler_job_main",
            replace_existing=True,
            max_instances=1,
            coalesce=True,
        )

    scheduler.start()
    _demand_scheduler = scheduler


@app.get("/demand/task/{task_id}/status")
def demand_task_status(task_id: int, max_log_chars: int = 2000):
    row = mysql_db.select_one(
        "demand_task",
        columns="id, execution_id, name, platform, status, log",
        where="id = %s",
        where_params=(task_id,),
    )
    if not row:
        raise HTTPException(status_code=404, detail="task not found")

    status = int(row.get("status") or 0)
    status_map = {0: "running", 1: "completed", 2: "failed"}

    log_text = row.get("log") or ""
    if max_log_chars and isinstance(log_text, str) and len(log_text) > max_log_chars:
        log_text = log_text[:max_log_chars] + "...(truncated)"

    execution_id = row.get("execution_id")
    final_text: Optional[str] = None
    if status == 1 and execution_id:
        try:
            result_path = Path(__file__).parent / "output" / str(execution_id) / "result.txt"
            if result_path.exists():
                with open(result_path, "r", encoding="utf-8") as f:
                    final_text = f.read()
        except Exception:
            final_text = None

    return {
        "task_id": task_id,
        "execution_id": execution_id,
        "name": row.get("name"),
        "platform": row.get("platform"),
        "status": status,
        "status_text": status_map.get(status, "unknown"),
        "final_text": final_text,
        "log": log_text,
    }


@app.get("/demand/tasks")
def demand_tasks(
        status: Optional[int] = None,
        name: Optional[str] = None,
        platform_type: Optional[str] = None,
        page: int = 1,
        page_size: int = 20,
):
    where_parts: list[str] = []
    where_params: list = []

    if status is not None:
        status_int = int(status)
        if status_int not in (0, 1, 2):
            raise HTTPException(status_code=400, detail="status 必须为 0/1/2")
        where_parts.append("status = %s")
        where_params.append(status_int)

    if name:
        name_str = str(name).strip()
        if name_str:
            # 支持模糊匹配：根据需求名称字段（varchar(32)）
            where_parts.append("name LIKE %s")
            where_params.append(f"%{name_str}%")

    if platform_type:
        platform_str = str(platform_type).strip()
        if platform_str:
            where_parts.append("platform = %s")
            where_params.append(platform_str)

    where = " AND ".join(where_parts)
    params = tuple(where_params) if where_params else None

    data = mysql_db.paginate(
        "demand_task",
        page=page,
        page_size=page_size,
        columns="id, execution_id, name, platform, status, create_time, update_time",
        where=where,
        where_params=params,
        order_by="id DESC",
    )

    # 返回分页结构（data + pagination），便于前端直接展示
    return data


def run_server():
    import uvicorn

    uvicorn.run(app, host="0.0.0.0", port=7000)


if __name__ == "__main__":
    run_server()