market_sync/app/tasks/task_kline_5min.py

"""同步任务：5 分钟 K 线（mairui）。

设计要点：
  1) 启动时**一次 SQL** 拿全表 {stock_code: max(bar_time)} 快照
  2) 内存里给每只股票算 (start, end, windows_needed)
       - DB 没数据 → start = mairui 历史深度起点 (2023-06-14)
       - DB 有数据 → start = max(2023-06-14, max(bar_time) - 2 天)
       - 兜底：start >= end → 跳过这只
  3) 输出"计划"（全量/增量分类、预估请求数、预估耗时）再开始执行
  4) 5 worker × 1 RPS/worker = 5 RPS 总（守住 mairui 上限）
  5) 窗口大小 1 年（mairui 一次最多 ~11600 条）

akshare 时代写法：start 写死 2020-01-01 + 4 天窗口，137 小时跑完全市场。
本版：增量时通常只补最近 1-2 天，~30 分钟。
"""
from __future__ import annotations

import os
import time
from concurrent.futures import ThreadPoolExecutor, as_completed
from datetime import datetime, timedelta, timezone
from typing import Any, Optional

from app.core.db import ops as db_ops
from app.core.datasource.base import registry as ds_registry
from app.core.datasource.registry import is_source_ready
from app.core.datasource.utils import is_a_share_code, to_code6, to_hermes
from app.core.sync.base import SyncTask
from app.core.sync.registry import mark_sync_blocked
from app.core.utils.logging import get_logger

logger = get_logger("sync.kline_5min")

# mairui 5min K 实测历史深度起点（实测 2023-06-14 才有数据，更早就 0）
MAIRUI_5MIN_FLOOR = datetime(2023, 6, 14, tzinfo=timezone.utc)
# 单次请求窗口大小：1 年（mairui 单次最多 ~11640 条 5min）
WINDOW_DAYS = 365
# 增量时往前多取的天数（防交易日历边界漏当天）
INCREMENT_OVERLAP_DAYS = 2


class SyncKline5Min(SyncTask):
    dataset_id = "kline_5min"

    def _plan(
        self,
        stock_codes: list[str],
        end: datetime,
        snapshots: dict[str, Optional[datetime]],
    ) -> list[tuple[str, datetime, datetime]]:
        """为每只股票算 (start, end) — 已排除 start >= end 的"无需同步"。

        Returns: [(code6, start, end), ...]
        """
        plans = []
        for code6 in stock_codes:
            latest = snapshots.get(code6)
            if latest is None:
                # DB 里没数据 → 全量从 mairui 历史深度起点
                start = MAIRUI_5MIN_FLOOR
            else:
                # DB 有数据 → 增量：从 max(bar_time) - overlap 到 end
                start = max(MAIRUI_5MIN_FLOOR, latest - timedelta(days=INCREMENT_OVERLAP_DAYS))
            if start < end:
                plans.append((code6, start, end))
        return plans

    def _run(
        self,
        *,
        trigger_source: str = "manual",
        codes: list[str] | None = None,
        max_workers: int = 5,
        **kwargs,
    ) -> dict[str, Any]:
        primary = ds_registry.get("datasource_mairui")
        if primary is None:
            return {"status": "error", "message": "5min 数据源 mairui 未注册"}
        ok, reason = is_source_ready(primary.key)
        if not ok:
            mark_sync_blocked(self.dataset_id, message=f"{primary.key} 未就绪: {reason}")
            return {"status": "blocked", "message": f"{primary.key} 未就绪: {reason}"}

        if codes:
            stock_codes = [to_code6(c) for c in codes]
        else:
            stock_codes = [c for c in db_ops.iter_stock_codes(active_only=True) if is_a_share_code(c)]
        limit_raw = os.environ.get("MARKET_DATA_STOCK_LIMIT", "").strip()
        if limit_raw.isdigit() and int(limit_raw) > 0:
            stock_codes = stock_codes[: int(limit_raw)]

        if not stock_codes:
            return {"status": "error", "message": "无股票代码"}

        # ── 1) 一次 SQL 拿全表快照 ──
        logger.info(f"[5min] 读取 DB 快照 ({len(stock_codes)} 只待查)…")
        snapshots = db_ops.get_kline_5min_snapshots()
        logger.info(f"[5min] DB 里已有 {len(snapshots)} 只的 kline_5min 数据")

        # ── 2) 内存算计划 ──
        # PG TIMESTAMPTZ 是 timezone-aware，所以 end 也必须 tz-aware 否则比较会炸
        end = datetime.now(timezone.utc)
        plans = self._plan(stock_codes, end, snapshots)
        if not plans:
            msg = f"5min K线 — 全部 {len(stock_codes)} 只都已最新（无需同步）"
            logger.info(msg)
            return {"status": "ok", "message": msg, "ok": 0, "fail": 0, "rows": 0, "elapsed_sec": 0.0}

        # 分类统计
        full_sync = [p for p in plans if snapshots.get(p[0]) is None]
        incr_sync = [p for p in plans if snapshots.get(p[0]) is not None]
        total_windows = sum(
            max(1, (p[2] - p[1]).days // WINDOW_DAYS + 1)
            for p in plans
        )
        # mairui 5 RPS 上限（5 worker × 1 RPS）
        est_sec = total_windows / 5.0
        logger.info(
            f"[5min] 计划: 总 {len(plans)} 只 (全量 {len(full_sync)} / 增量 {len(incr_sync)})，"
            f"总请求 {total_windows}，按 5 RPS 预估 {est_sec:.0f}s ({est_sec/60:.1f}min)"
        )
        self._progress(
            message=f"计划: {len(plans)} 只 (全量 {len(full_sync)}/增量 {len(incr_sync)}) "
                    f"约 {total_windows} 个请求, 预估 {est_sec/60:.1f}min",
            current=0, total=len(plans), current_step="planning",
        )

        # ── 3) 执行 ──
        t0 = time.time()
        ok_cnt = fail_cnt = 0
        rows_total = 0
        with ThreadPoolExecutor(max_workers=max_workers) as pool:
            futures = {pool.submit(self._sync_one, primary, p[0], p[1], p[2]): p[0] for p in plans}
            for i, future in enumerate(as_completed(futures), 1):
                c6 = futures[future]
                try:
                    res = future.result()
                    if res["status"] == "ok":
                        ok_cnt += 1
                        rows_total += res["rows"]
                    else:
                        fail_cnt += 1
                        if res.get("error"):
                            logger.warning(f"[5min {c6}] {res['error']}")
                except Exception as e:
                    fail_cnt += 1
                    logger.warning(f"[5min {c6}] {e}")
                if i % 20 == 0 or i == len(plans):
                    self._progress(
                        message=f"5min 进度 {i}/{len(plans)} OK:{ok_cnt} FAIL:{fail_cnt}",
                        current=i, total=len(plans), current_step=c6,
                    )
        elapsed = round(time.time() - t0, 1)
        msg = (
            f"5min K线 {ok_cnt}成 {fail_cnt}败 共{rows_total}行, "
            f"全量{len(full_sync)}/增量{len(incr_sync)}, {elapsed}s"
        )
        return {
            "status": "ok" if fail_cnt == 0 else "warning",
            "message": msg,
            "ok": ok_cnt, "fail": fail_cnt, "rows": rows_total,
            "full_sync": len(full_sync), "incr_sync": len(incr_sync),
            "total_requests": total_windows, "elapsed_sec": elapsed,
        }

    def _sync_one(self, primary, code6: str, start: datetime, end: datetime) -> dict:
        all_rows: list[dict] = []
        try:
            cur = start
            while cur < end:
                w_end = min(cur + timedelta(days=WINDOW_DAYS), end)
                df = primary.fetch_kline_5min(
                    code6,
                    cur.strftime("%Y-%m-%d"),
                    w_end.strftime("%Y-%m-%d"),
                )
                if df is not None and not df.empty:
                    hermes = to_hermes(code6)
                    for _, r in df.iterrows():
                        bar_time = r["bar_time"]
                        all_rows.append({
                            "stock_code": hermes,
                            "bar_time": bar_time.strftime("%Y-%m-%d %H:%M:%S")
                                      if hasattr(bar_time, "strftime") else str(bar_time),
                            "open": float(r.get("open") or 0),
                            "high": float(r.get("high") or 0),
                            "low": float(r.get("low") or 0),
                            "close": float(r.get("close") or 0),
                            "volume": float(r.get("volume") or 0),
                            "amount": float(r.get("amount") or 0),
                            "turnover_rate": float(r.get("turnover_rate") or 0),
                        })
                cur = w_end + timedelta(days=1)
        except Exception as e:
            return {"status": "fail", "error": str(e)}

        if not all_rows:
            return {"status": "fail", "error": "no data"}
        try:
            CHUNK = 5000
            for i in range(0, len(all_rows), CHUNK):
                db_ops.upsert_kline_5min(all_rows[i: i + CHUNK])
        except Exception as e:
            return {"status": "fail", "error": f"db write: {e}"}
        return {"status": "ok", "rows": len(all_rows)}