feat(HEL-421): 回补历史日历和指数并标记区间不完整

Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: multica-agent <github@multica.ai>
This commit is contained in:
总工
2026-09-02 22:16:05 +08:00
co-authored by Cursor multica-agent
parent 25ff6bbe06
commit a836cda1b2
18 changed files with 715 additions and 33 deletions
+168 -13
View File
@@ -7,7 +7,7 @@ from datetime import timedelta
from typing import Any
from datahub.adapters.base import AdapterError
from datahub.adapters.tushare import DEFAULT_INDEX_CODES, TushareAdapter
from datahub.adapters.tushare import DEFAULT_INDEX_CODES, WEBSITE_INDEX_CODES, TushareAdapter
from datahub.db import DATASET_TABLES, HubDB
from datahub.governance.circuit import CircuitBreaker
from datahub.governance.ratelimit import TokenBucket
@@ -141,11 +141,22 @@ class Pipeline:
seq = int((row or {}).get("n") or 0) + 1
return f"{trade_date}-{dataset}-{seq:03d}"
def ingest_reference(self, trade_date: str | None = None) -> dict[str, Any]:
"""Refresh trade calendar (window) and stock master. Not versioned by batch."""
def ingest_reference(
self,
trade_date: str | None = None,
start: str | None = None,
end: str | None = None,
) -> dict[str, Any]:
"""Refresh trade calendar and stock master. Not versioned by batch.
Calendar defaults to 2016-01-01 through today+30 so a 5-year website
query is not silently truncated. UPSERT makes repeats safe.
"""
day = yyyymmdd(trade_date or self.clock())
start = add_days(day, -400)
end = add_days(day, 30)
start = yyyymmdd(start or self.settings.calendar_start)
end = yyyymmdd(end or add_days(day, 30))
if start > end:
start, end = end, start
calendar = self.adapter.normalize(
"calendar",
self._guarded_fetch("calendar", {"exchange": "SSE", "start_date": start, "end_date": end}),
@@ -180,9 +191,150 @@ class Pipeline:
row.get("list_date"), fetched_at,
),
)
return {"calendar": len(calendar), "stocks": len(stocks), "trade_date": day}
return {
"calendar": len(calendar),
"stocks": len(stocks),
"trade_date": day,
"calendar_from": start,
"calendar_to": end,
}
def run_dataset(self, dataset: str, trade_date: str, attempts: int | None = None) -> dict[str, Any]:
def open_trade_dates(self, end: str, limit: int) -> list[str]:
end = yyyymmdd(end)
rows = self.db.fetchall(
"""
SELECT cal_date FROM trade_calendar
WHERE exchange = 'SSE' AND is_open = 1 AND cal_date <= ?
ORDER BY cal_date DESC
LIMIT ?
""",
(end, max(1, int(limit))),
)
return sorted(str(row["cal_date"]) for row in rows)
def backfill_history(
self,
trade_date: str | None = None,
calendar_start: str | None = None,
index_days: int | None = None,
codes: tuple[str, ...] | None = None,
force: bool = False,
) -> dict[str, Any]:
"""Idempotent calendar + website-index history backfill."""
day = yyyymmdd(trade_date or self.clock())
calendar = self.ingest_reference(day, start=calendar_start)
index = self.backfill_index_history(
end_date=day,
trading_days=index_days,
codes=codes,
force=force,
)
return {"calendar": calendar, "index_daily": index, "ok": bool(index.get("ok"))}
def backfill_index_history(
self,
end_date: str | None = None,
trading_days: int | None = None,
codes: tuple[str, ...] | None = None,
force: bool = False,
) -> dict[str, Any]:
"""Incrementally publish official index bars for website index codes.
One range fetch per code, then per-day publish. Already published dates
are skipped unless ``force``. Failures are recorded and do not roll back
successful days.
"""
end = yyyymmdd(end_date or self.clock())
limit = int(trading_days or self.settings.index_history_trading_days)
codes = tuple(codes or WEBSITE_INDEX_CODES)
open_dates = self.open_trade_dates(end, limit)
if not open_dates:
return {
"start": None,
"end": end,
"codes": list(codes),
"requested_days": 0,
"published": [],
"skipped": [],
"failed": [{"error": "calendar has no open dates on or before end"}],
"ok": False,
}
start = open_dates[0]
complete_dates = set() if force else self._index_dates_with_all_codes(start, end, codes)
targets = [day for day in open_dates if day not in complete_dates]
skipped = [day for day in open_dates if day in complete_dates]
by_date: dict[str, list[dict[str, Any]]] = {day: [] for day in targets}
failed: list[dict[str, Any]] = []
for ts_code in codes:
try:
raw = retry_call(
lambda code=ts_code: self._guarded_fetch(
"index_daily",
{"ts_code": code, "start_date": start, "end_date": end},
),
attempts=self.settings.max_publish_attempts,
base_delay=0.05,
sleeper=lambda _d: time.sleep(_d),
)
for row in self.adapter.normalize("index_daily", raw):
day = str(row.get("trade_date") or "")
if day in by_date:
by_date[day].append(row)
except Exception as exc:
failed.append({"ts_code": ts_code, "error": str(exc)})
published: list[dict[str, Any]] = []
for day in targets:
rows = by_date.get(day) or []
try:
result = self.run_dataset("index_daily", day, prepared_rows=rows)
published.append(
{
"trade_date": day,
"batch_id": result["batch_id"],
"rows": result["rows"],
"state": result["state"],
}
)
except Exception as exc:
failed.append({"trade_date": day, "error": str(exc), "rows": len(rows)})
return {
"start": start,
"end": end,
"codes": list(codes),
"requested_days": len(open_dates),
"published": published,
"skipped": skipped,
"failed": failed,
"ok": not failed,
}
def _index_dates_with_all_codes(self, start: str, end: str, codes: tuple[str, ...]) -> set[str]:
pubs = self.db.fetchall(
"""
SELECT trade_date, active_batch FROM publications
WHERE dataset = 'index_daily' AND trade_date >= ? AND trade_date <= ?
""",
(start, end),
)
needed = set(codes)
complete: set[str] = set()
for pub in pubs:
rows = self.db.fetchall(
"SELECT DISTINCT ts_code FROM eod_index_bars WHERE trade_date = ? AND batch_id = ?",
(pub["trade_date"], pub["active_batch"]),
)
have = {str(row["ts_code"]) for row in rows}
if needed <= have:
complete.add(str(pub["trade_date"]))
return complete
def run_dataset(
self,
dataset: str,
trade_date: str,
attempts: int | None = None,
prepared_rows: list[dict[str, Any]] | None = None,
) -> dict[str, Any]:
trade_date = yyyymmdd(trade_date)
batch_id = self.next_batch_id(dataset, trade_date)
max_attempts = attempts or self.settings.max_publish_attempts
@@ -190,12 +342,15 @@ class Pipeline:
rows: list[dict[str, Any]] = []
try:
self._set_batch(batch_id, dataset, trade_date, "fetching", 1)
rows = retry_call(
lambda: self._fetch_dataset(dataset, trade_date),
attempts=max_attempts,
base_delay=0.05,
sleeper=lambda _d: None if attempts == 1 else time.sleep(_d),
)
if prepared_rows is None:
rows = retry_call(
lambda: self._fetch_dataset(dataset, trade_date),
attempts=max_attempts,
base_delay=0.05,
sleeper=lambda _d: None if attempts == 1 else time.sleep(_d),
)
else:
rows = list(prepared_rows)
self._stage(dataset, batch_id, rows)
self._set_batch(batch_id, dataset, trade_date, "staged", 1, rows_in=len(rows), rows_out=len(rows))
self._set_batch(batch_id, dataset, trade_date, "validating", 1)