gao
daef609e8b
- QmtBridgeSource: provides 增加 index_daily，新增 fetch_index_daily 方法 - task_kline_index: 优先级改为 qmt_bridge → mairui → sina 三级降级 - task_kline_5min: 优先用 qmt_bridge，不可用时降级 mairui - task_kline_daily: 优先级加入 qmt_bridge（首位） - config: 新增 qmt_bridge_url 配置项 - registry: qmt_bridge 注册信息同步更新 - docs: 新增 DATASETS_AND_SOURCES.md，完整说明数据集与数据源依赖关系 - AGENTS.md: 同步更新指数数据源描述
378 lines
16 KiB
Python
378 lines
16 KiB
Python
"""同步任务：全市场日 K 线。
|
||
|
||
数据流：
|
||
1. 选主源（按 priority 选第一个 is_available 的源）
|
||
2. 增量判断：读 kline_stock.MAX(trade_date) 决定每只股票 start
|
||
3. **Producer-Consumer 模式**：
|
||
- N 个 fetcher 线程并发拉 K 线，把 (code, rows) 塞到 Queue
|
||
- 1 个 writer 线程攒满 BATCH_SIZE 就 executemany + commit 一次
|
||
4. 更新 stocks.kline_synced_at
|
||
|
||
设计要点：
|
||
- 避免每只股票 commit（5000 只 = 5000 次 commit 太慢）
|
||
- 减少并发数（避免新浪限流 / MySQL 连接数爆）
|
||
- fetcher 间用 Queue 解耦，写库串行避免锁竞争
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
import os
|
||
import queue
|
||
import threading
|
||
import time
|
||
from concurrent.futures import ThreadPoolExecutor, as_completed, TimeoutError as FuturesTimeoutError
|
||
from datetime import datetime, timedelta
|
||
from typing import Any
|
||
|
||
import pandas as pd
|
||
|
||
from app.core.db import ops as db_ops
|
||
from app.core.datasource.base import registry as ds_registry
|
||
from app.core.datasource.utils import is_a_share_code, to_code6, to_hermes
|
||
from app.core.sync.base import SyncTask, _effective_sync_end
|
||
from app.core.utils.logging import get_logger
|
||
|
||
logger = get_logger("sync.kline_daily")
|
||
|
||
# 源优先级：QMT Bridge（本地） → 麦蕊 → 雪球 → 新浪
|
||
PRIMARY_PRIORITY = ["datasource_qmt_bridge", "datasource_mairui", "datasource_xueqiu", "datasource_xinlang"]
|
||
DEFAULT_START = "2018-01-01"
|
||
# 写库批量：攒够 BATCH_SIZE 行就 executemany 一次 + commit
|
||
BATCH_SIZE = 2000
|
||
# fetcher 并发：实测单线程最稳（5000 只 × 0.2s = 17 分钟，足够）
|
||
# 雪球 token 限速 10 RPS，单线程 0.1s/只 已经打满。多 worker 会触发风控 hang
|
||
DEFAULT_FETCH_WORKERS = 1
|
||
# 2026-07-08 教训: 数据源层虽然加了 fetch timeout (20s), task 层也加硬上限兜底,
|
||
# 防止数据源 SDK 升级后又被绕过。60s 对应 xueqiu/mairui 的 15-20s 上限 + DB upsert
|
||
# / DataFrame 处理余量 + 7月9日 mairui 间歇慢请求的容忍窗口。
|
||
FETCH_HARD_TIMEOUT = 180.0
|
||
|
||
|
||
class SyncKlineDaily(SyncTask):
|
||
dataset_id = "kline_daily"
|
||
|
||
def _run(
|
||
self,
|
||
*,
|
||
trigger_source: str = "manual",
|
||
start: str | None = None,
|
||
end: str | None = None,
|
||
codes: list[str] | None = None,
|
||
incremental: bool = True,
|
||
max_workers: int = DEFAULT_FETCH_WORKERS,
|
||
**kwargs,
|
||
) -> dict[str, Any]:
|
||
# 1. 获取所有可用源（按优先级），支持 per-stock fallback
|
||
sources = self._get_available_sources()
|
||
if not sources:
|
||
return {"status": "error", "message": "所有 K 线数据源均不可用"}
|
||
source_names = ", ".join(s.name for s in sources)
|
||
|
||
self._progress(
|
||
message=f"使用 [{source_names}] 同步日K线 (fetcher={max_workers}, batch={BATCH_SIZE})...",
|
||
current_step=sources[0].key,
|
||
)
|
||
|
||
# 2. 拉股票列表
|
||
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（stocks 表为空？）"}
|
||
|
||
# 3. 增量判断
|
||
_end = end or _effective_sync_end()
|
||
end_date_obj = datetime.strptime(_end, "%Y-%m-%d").date()
|
||
jobs: list[tuple[str, str]] = []
|
||
if incremental and not start:
|
||
for c6 in stock_codes:
|
||
last = db_ops.get_stock_kline_max_date(c6)
|
||
if last is None:
|
||
fetch_start = DEFAULT_START
|
||
elif datetime.strptime(last, "%Y-%m-%d").date() >= end_date_obj:
|
||
continue
|
||
else:
|
||
fetch_start = (datetime.strptime(last, "%Y-%m-%d").date() - timedelta(days=5)).strftime("%Y-%m-%d")
|
||
jobs.append((c6, fetch_start))
|
||
else:
|
||
fetch_start = start or DEFAULT_START
|
||
jobs = [(c, fetch_start) for c in stock_codes]
|
||
|
||
total = len(jobs)
|
||
skipped = len(stock_codes) - total
|
||
self._progress(
|
||
message=f"开始同步 {total} 只股票 (跳过 {skipped} 只已最新)",
|
||
current=0, total=total,
|
||
)
|
||
|
||
# 4. 并发 fetch + 攒 batch 写库（线程安全）
|
||
from concurrent.futures import ThreadPoolExecutor, as_completed
|
||
t0 = time.time()
|
||
batch_lock = threading.Lock()
|
||
batch_rows: list[dict] = []
|
||
pending_codes: dict[str, str] = {}
|
||
ok_lock = threading.Lock()
|
||
ok_cnt = [0]
|
||
fallback_ok_cnt = [0]
|
||
fail_cnt = [0]
|
||
rows_cnt = [0]
|
||
|
||
def _flush():
|
||
with batch_lock:
|
||
if not batch_rows:
|
||
return
|
||
rows_to_write = list(batch_rows)
|
||
codes_to_update = dict(pending_codes)
|
||
batch_rows.clear()
|
||
pending_codes.clear()
|
||
try:
|
||
db_ops.upsert_kline_stock(rows_to_write)
|
||
except Exception as e:
|
||
logger.error(f"batch upsert 失败 ({len(rows_to_write)} 行): {e}")
|
||
return
|
||
for c6, latest in codes_to_update.items():
|
||
try:
|
||
db_ops.update_stock_kline_synced_at(c6, latest)
|
||
except Exception as e:
|
||
logger.warning(f"update synced_at 失败 {c6}: {e}")
|
||
|
||
def _rows_from_df(code6: str, df: pd.DataFrame) -> tuple[list[dict], str]:
|
||
hermes = to_hermes(code6)
|
||
rows = []
|
||
for _, r in df.iterrows():
|
||
td = r["trade_date"]
|
||
rows.append({
|
||
"stock_code": hermes,
|
||
"trade_date": td.strftime("%Y-%m-%d") if hasattr(td, "strftime") else str(td)[:10],
|
||
"open": float(r["open"]),
|
||
"high": float(r["high"]),
|
||
"low": float(r["low"]),
|
||
"close": float(r["close"]),
|
||
"volume": float(r["volume"]),
|
||
})
|
||
latest = df["trade_date"].max()
|
||
latest = latest.strftime("%Y-%m-%d") if hasattr(latest, "strftime") else str(latest)[:10]
|
||
return (rows, latest)
|
||
|
||
def _fetch_one(code6: str, fetch_start: str) -> tuple[str, list[dict] | None, str | None, bool]:
|
||
errors = []
|
||
for src in sources:
|
||
try:
|
||
df = src.fetch_kline_daily(code6, fetch_start, _end)
|
||
except Exception as e:
|
||
errors.append(f"{src.key}: {e}")
|
||
continue
|
||
if df is None or df.empty:
|
||
errors.append(f"{src.key}: no data")
|
||
continue
|
||
rows, latest = _rows_from_df(code6, df)
|
||
is_fallback = src is not sources[0]
|
||
return (code6, rows, latest, is_fallback)
|
||
return (code6, None, "; ".join(errors)[:200], False)
|
||
|
||
total = len(jobs)
|
||
# 不用 `with ThreadPoolExecutor(...) as pool:` — 它的 __exit__ 默认 wait=True,
|
||
# 一旦某个 worker 卡在 xueqiu IO,主线程会在 as_completed 触发 timeout 后
|
||
# 仍被 __exit__ 阻塞等 worker 退出 → 进程挂死(2026-07-09 13:21 教训)。
|
||
# 改成手动管理 + shutdown(wait=False),主线程能立刻退出。
|
||
pool = ThreadPoolExecutor(max_workers=max_workers)
|
||
futures = {pool.submit(_fetch_one, c6, fs): c6 for c6, fs in jobs}
|
||
done_cnt = 0
|
||
try:
|
||
for future in as_completed(futures, timeout=FETCH_HARD_TIMEOUT):
|
||
done_cnt += 1
|
||
code6 = futures[future]
|
||
try:
|
||
code6_r, rows, latest, is_fallback = future.result(timeout=0.1)
|
||
except FuturesTimeoutError:
|
||
with ok_lock:
|
||
fail_cnt[0] += 1
|
||
logger.warning(f"[kline {code6}] 内部 race timeout, 记 fail")
|
||
continue
|
||
except Exception as e:
|
||
with ok_lock:
|
||
fail_cnt[0] += 1
|
||
logger.warning(f"[kline {code6}] 异常: {e}")
|
||
continue
|
||
if rows is None:
|
||
with ok_lock:
|
||
fail_cnt[0] += 1
|
||
logger.warning(f"[kline {code6}] {latest}")
|
||
continue
|
||
if is_fallback:
|
||
with ok_lock:
|
||
fallback_ok_cnt[0] += 1
|
||
with batch_lock:
|
||
batch_rows.extend(rows)
|
||
pending_codes[code6] = latest
|
||
cur_size = len(batch_rows)
|
||
with ok_lock:
|
||
ok_cnt[0] += 1
|
||
rows_cnt[0] += len(rows)
|
||
|
||
if cur_size >= BATCH_SIZE:
|
||
_flush()
|
||
|
||
if done_cnt % 50 == 0 or done_cnt == total:
|
||
elapsed = time.time() - t0
|
||
rate = done_cnt / elapsed if elapsed > 0 else 0
|
||
self._progress(
|
||
message=f"进度 {done_cnt}/{total} OK:{ok_cnt[0]} FAIL:{fail_cnt[0]} {rate:.1f}只/秒",
|
||
current=done_cnt, total=total, current_step=code6,
|
||
)
|
||
except FuturesTimeoutError:
|
||
# 30s 内没 future 完成 → 所有 worker 都被卡死(数据源层 timeout 兜底失败)
|
||
# 取消 pending futures,标 fail。shutdown(wait=False) 让主线程立刻返回,
|
||
# 不被卡住的 worker 拖死。
|
||
# 2026-07-11 修复: 防御性 cancel,避免异常处理链里变量被遮蔽导致
|
||
# AttributeError: 'str' object has no attribute 'cancel'。
|
||
stuck = [(f, c6) for f, c6 in futures.items() if not f.done()]
|
||
logger.error(
|
||
"所有 fetcher 卡死 (>%ss), %d 只股票未完成, 标 fail 后退出主循环",
|
||
FETCH_HARD_TIMEOUT, len(stuck),
|
||
)
|
||
with ok_lock:
|
||
fail_cnt[0] += len(stuck)
|
||
for f, c6 in stuck:
|
||
try:
|
||
if hasattr(f, "cancel"):
|
||
f.cancel()
|
||
except Exception as cancel_err:
|
||
logger.warning(f"[kline] cancel future for {c6} 失败: {cancel_err}")
|
||
try:
|
||
pool.shutdown(wait=False)
|
||
except Exception as shutdown_err:
|
||
logger.warning(f"[kline] pool.shutdown(wait=False) 失败: {shutdown_err}")
|
||
|
||
# ── 补偿：冷却后重入队列 ──
|
||
# 雪球/mairui 偶发全 hang（服务端限流/网络抖动），等 30s 再试一次。
|
||
# 单线程拉取场景下，卡住通常意味着当前这只股票请求 hang 死；
|
||
# 冷却后把未完成的股票重新排队重试，救回临时故障。
|
||
if stuck:
|
||
cooldown = 30
|
||
logger.warning(f"[kline] 冷却 {cooldown}s 后重试 {len(stuck)} 只…")
|
||
time.sleep(cooldown)
|
||
retry_ok = 0
|
||
retry_fb = 0
|
||
retry_rows = 0
|
||
retry_pool = ThreadPoolExecutor(max_workers=max_workers)
|
||
jobs_dict = {c6: fs for c6, fs in jobs}
|
||
retry_futs = {}
|
||
for _, c6 in stuck:
|
||
if c6 in jobs_dict:
|
||
retry_futs[retry_pool.submit(_fetch_one, c6, jobs_dict[c6])] = c6
|
||
if retry_futs:
|
||
try:
|
||
for fut in as_completed(retry_futs, timeout=FETCH_HARD_TIMEOUT):
|
||
c6 = retry_futs[fut]
|
||
try:
|
||
code6_r, rows, latest, is_fallback = fut.result(timeout=0.1)
|
||
except Exception:
|
||
logger.warning(f"[kline {c6}] 重试异常")
|
||
continue
|
||
if rows is None:
|
||
logger.warning(f"[kline {c6}] 重试仍失败: {latest}")
|
||
continue
|
||
with batch_lock:
|
||
batch_rows.extend(rows)
|
||
pending_codes[c6] = latest
|
||
cur_size = len(batch_rows)
|
||
with ok_lock:
|
||
retry_ok += 1
|
||
retry_rows += len(rows)
|
||
if is_fallback:
|
||
retry_fb += 1
|
||
if cur_size >= BATCH_SIZE:
|
||
_flush()
|
||
except FuturesTimeoutError:
|
||
logger.error(
|
||
"[kline] 重试仍卡死 (>%ss), %d 只股票放弃",
|
||
FETCH_HARD_TIMEOUT, sum(1 for f in retry_futs if not f.done()),
|
||
)
|
||
try:
|
||
retry_pool.shutdown(wait=False)
|
||
except Exception:
|
||
pass
|
||
with ok_lock:
|
||
ok_cnt[0] += retry_ok
|
||
fail_cnt[0] -= retry_ok
|
||
fallback_ok_cnt[0] += retry_fb
|
||
rows_cnt[0] += retry_rows
|
||
logger.warning(
|
||
f"[kline] 重试结果: {retry_ok}成 "
|
||
f"({retry_fb}只备胎) 共{retry_rows}行, 救回 {retry_ok} 只"
|
||
)
|
||
else:
|
||
pool.shutdown(wait=True)
|
||
|
||
# 收尾 flush
|
||
_flush()
|
||
|
||
elapsed = round(time.time() - t0, 1)
|
||
fb = fallback_ok_cnt[0]
|
||
fb_info = f" ({fb}只备胎)" if fb else ""
|
||
msg = f"日K线 {ok_cnt[0]}成 {fail_cnt[0]}败 {skipped}跳 共{rows_cnt[0]}行{fb_info}, {elapsed}s"
|
||
return {
|
||
"status": "ok" if fail_cnt[0] == 0 else "warning",
|
||
"message": msg,
|
||
"ok": ok_cnt[0],
|
||
"fail": fail_cnt[0],
|
||
"fallback_ok": fb,
|
||
"skip": skipped,
|
||
"rows": rows_cnt[0],
|
||
"elapsed_sec": elapsed,
|
||
"primary_source": sources[0].key if sources else "",
|
||
"sources": [s.key for s in sources],
|
||
}
|
||
|
||
def _get_available_sources(self):
|
||
from app.core.datasource.registry import is_source_ready, run_health_check, get_health_status
|
||
if not get_health_status():
|
||
run_health_check()
|
||
available = []
|
||
for key in PRIMARY_PRIORITY:
|
||
ok, _ = is_source_ready(key)
|
||
if ok:
|
||
available.append(ds_registry.get(key))
|
||
return available
|
||
|
||
def _sync_one(self, primary, code6: str, start: str, end: str) -> dict:
|
||
"""保留这个方法以兼容外部调用（已不用，但单只测试可能用到）"""
|
||
try:
|
||
df = primary.fetch_kline_daily(code6, start, end)
|
||
except Exception as e:
|
||
return {"status": "fail", "error": str(e)}
|
||
if df is None or df.empty:
|
||
return {"status": "fail", "error": "no data"}
|
||
hermes = to_hermes(code6)
|
||
rows = []
|
||
for _, r in df.iterrows():
|
||
td = r["trade_date"]
|
||
rows.append({
|
||
"stock_code": hermes,
|
||
"trade_date": td.strftime("%Y-%m-%d") if hasattr(td, "strftime") else str(td)[:10],
|
||
"open": float(r["open"]),
|
||
"high": float(r["high"]),
|
||
"low": float(r["low"]),
|
||
"close": float(r["close"]),
|
||
"volume": float(r["volume"]),
|
||
})
|
||
try:
|
||
db_ops.upsert_kline_stock(rows)
|
||
except Exception as e:
|
||
return {"status": "fail", "error": f"db write: {e}"}
|
||
if rows:
|
||
latest = max(r["trade_date"] for r in rows)
|
||
try:
|
||
db_ops.update_stock_kline_synced_at(code6, latest)
|
||
except Exception:
|
||
pass
|
||
return {"status": "ok", "rows": len(rows)}
|