feat(HEL-435): 盘后未出数时晚间自动重试并提供安全补跑

Co-authored-by: multica-agent <github@multica.ai>

xiaobai-datahub/datahub/scheduler.py

@@ -2,7 +2,7 @@ from __future__ import annotations
 
 import threading
 from collections.abc import Callable
-from datetime import datetime, time
+from datetime import datetime, time, timedelta
 from typing import Any
 
 from datahub.db import HubDB
@@ -14,6 +14,8 @@ LOGGER = get_logger()
 
 JobFn = Callable[[str], Any]
 
+EOD_JOB_IDS = {"eod_a", "eod_b", "eod_retry"}
+
 
 def is_open_day(db: HubDB, day: str) -> bool:
     row = db.fetchone(
@@ -25,8 +27,19 @@ def is_open_day(db: HubDB, day: str) -> bool:
     return int(row["is_open"]) == 1
 
 
+def _hhmm(value: str) -> time:
+    return datetime.strptime(value, "%H:%M").time()
+
+
 class Scheduler:
-    """Calendar-driven in-process scheduler. Non-trading days skip EOD fetches."""
+    """Calendar-driven in-process scheduler. Non-trading days skip EOD fetches.
+
+    EOD datasets that failed to publish (e.g. upstream not ready at 15:05)
+    are retried automatically every ``eod_retry_interval_minutes`` between
+    ``eod_retry_start`` and ``eod_retry_cutoff``. Progress is persisted in
+    ``eod_progress`` so a container restart catches up instead of waiting
+    for the next day, and completed days are never re-fetched.
+    """
 
     def __init__(self, db: HubDB, pipeline: Pipeline, jobs: dict[str, JobFn] | None = None) -> None:
         self.db = db
@@ -35,6 +48,7 @@ class Scheduler:
             "precheck": self._precheck,
             "eod_a": self._eod_a,
             "eod_b": self._eod_b,
+            "eod_retry": self._eod_retry,
             "cleanup": self._cleanup,
             "backup": self._backup,
             "history_backfill": self._history_backfill,
@@ -42,6 +56,7 @@ class Scheduler:
         self._stop = threading.Event()
         self._thread: threading.Thread | None = None
         self._fired: set[tuple[str, str, str]] = set()
+        self._eod_lock = threading.Lock()
 
     def start(self, interval_seconds: float = 30.0) -> None:
         if self._thread and self._thread.is_alive():
@@ -63,9 +78,9 @@ class Scheduler:
             self._thread.join(timeout)
 
     def tick(self, clock: datetime | None = None) -> list[str]:
-        now = clock or now_shanghai()
+        now = now_shanghai(clock)
         day = yyyymmdd(now)
-        current = now.timetz() if False else now.time()
+        current = now.time()
         ran: list[str] = []
         plan = [
             ("precheck", time(8, 45)),
@@ -85,14 +100,183 @@ class Scheduler:
                 self._fired.add(key)
                 continue
             self._fired.add(key)
-            self.run_job(job_id, day)
+            if job_id in {"eod_a", "eod_b"}:
+                # Record the attempt before running: even a crash must not
+                # hide that today's first EOD try already happened.
+                self._record_eod_attempt(day, now)
+            try:
+                self.run_job(job_id, day)
+            except Exception:
+                if job_id not in {"eod_a", "eod_b"}:
+                    raise
+                # Keep the tick alive; evening retries take over.
+                LOGGER.exception("scheduled job %s failed for %s", job_id, day)
             ran.append(job_id)
+            if job_id in {"eod_a", "eod_b"}:
+                self._settle_eod(day)
+        ran.extend(self._eod_retry_tick(now, day, open_day))
         return ran
 
+    # ------------------------------------------------------------------
+    # EOD retry window
+    # ------------------------------------------------------------------
+
+    def _eod_retry_tick(self, now: datetime, day: str, open_day: bool) -> list[str]:
+        if not open_day:
+            return []
+        settings = self.pipeline.settings
+        current = now.time()
+        start = _hhmm(settings.eod_retry_start)
+        cutoff = _hhmm(settings.eod_retry_cutoff)
+        interval = timedelta(minutes=settings.eod_retry_interval_minutes)
+        missing = self.pipeline.missing_official_datasets(day)
+        row = self.eod_progress(day)
+
+        if not missing:
+            if row is None or row["state"] != "done":
+                self._save_eod_progress(day, state="done", finished_at=isoformat(now))
+            return []
+        if current < start:
+            return []
+        if row and row["state"] == "cutoff_failed":
+            return []
+        if current >= cutoff:
+            detail = "截止时间已到，缺失数据集: " + ",".join(missing)
+            self._save_eod_progress(day, state="cutoff_failed", finished_at=isoformat(now), detail=detail)
+            with self.db.write() as connection:
+                connection.execute(
+                    "INSERT INTO job_runs(job_id, state, started_at, finished_at, error, attempt, detail)"
+                    " VALUES ('eod_retry','failed',?,?,?,?,?)",
+                    (isoformat(now), isoformat(now), detail, int((row or {}).get("attempts") or 0), "eod cutoff reached"),
+                )
+            LOGGER.warning(
+                "eod retry window closed without data",
+                extra={"hub": {"trade_date": day, "missing": missing, "reason": "eod_cutoff"}},
+            )
+            return []
+        last = None
+        if row and row["last_attempt_at"]:
+            try:
+                last = datetime.fromisoformat(str(row["last_attempt_at"]))
+            except ValueError:
+                last = None
+        if last is not None and now_shanghai(last).replace(tzinfo=None) + interval > now.replace(tzinfo=None):
+            return []
+        if "eod_retry" not in self.jobs:
+            return []
+        self._record_eod_attempt(day, now)
+        ran = []
+        try:
+            self.run_job("eod_retry", day)
+        except Exception:
+            # job_runs already carries the failure; the window keeps retrying.
+            LOGGER.warning("eod retry failed for %s", day, exc_info=True)
+        ran.append("eod_retry")
+        self._settle_eod(day)
+        return ran
+
+    def _settle_eod(self, day: str) -> None:
+        """Flip the day to done as soon as every official dataset is published."""
+        if not self.pipeline.missing_official_datasets(day):
+            row = self.eod_progress(day)
+            if row is None or row["state"] != "done":
+                self._save_eod_progress(day, state="done", finished_at=isoformat())
+
+    def eod_progress(self, day: str) -> dict[str, Any] | None:
+        return self.db.fetchone("SELECT * FROM eod_progress WHERE trade_date = ?", (day,))
+
+    def eod_status(self, trade_date: str | None = None, clock: datetime | None = None) -> dict[str, Any]:
+        """Human/admin facing view: 等待上游 / 下次重试 / 已成功 / 已截止失败."""
+        day = yyyymmdd(trade_date or now_shanghai(clock))
+        now = now_shanghai(clock)
+        row = self.eod_progress(day)
+        open_day = is_open_day(self.db, day)
+        missing = self.pipeline.missing_official_datasets(day)
+        if row and row["state"] == "done":
+            state = "done"
+        elif not open_day:
+            state = "closed_day"
+        elif not missing:
+            state = "done"
+        elif row and row["state"] == "cutoff_failed":
+            state = "cutoff_failed"
+        elif now.time() < _hhmm("15:05"):
+            state = "pending_first_attempt"
+        else:
+            state = "waiting_upstream"
+        return {
+            "trade_date": day,
+            "is_open_day": open_day,
+            "state": state,
+            "missing_datasets": missing,
+            "attempts": int((row or {}).get("attempts") or 0),
+            "last_attempt_at": (row or {}).get("last_attempt_at"),
+            "next_retry_at": (row or {}).get("next_retry_at") if state == "waiting_upstream" else None,
+            "finished_at": (row or {}).get("finished_at"),
+            "detail": (row or {}).get("detail"),
+        }
+
+    def _record_eod_attempt(self, day: str, now: datetime) -> None:
+        row = self.eod_progress(day)
+        attempts = int((row or {}).get("attempts") or 0) + 1
+        interval = self.pipeline.settings.eod_retry_interval_minutes
+        self._save_eod_progress(
+            day,
+            state="waiting_upstream",
+            attempts=attempts,
+            last_attempt_at=isoformat(now),
+            next_retry_at=isoformat(now + timedelta(minutes=interval)),
+        )
+
+    def _save_eod_progress(self, day: str, **fields: Any) -> None:
+        columns = [
+            "trade_date", "state", "attempts", "last_attempt_at",
+            "next_retry_at", "finished_at", "detail", "updated_at",
+        ]
+        with self.db.write() as connection:
+            existing = connection.execute(
+                "SELECT trade_date FROM eod_progress WHERE trade_date = ?",
+                (day,),
+            ).fetchone()
+            if existing is None:
+                payload = {name: None for name in columns}
+                payload.update({"trade_date": day, "state": "waiting_upstream", "attempts": 0})
+                payload.update(fields)
+                payload["updated_at"] = isoformat()
+                placeholders = ",".join("?" for _ in columns)
+                connection.execute(
+                    f"INSERT INTO eod_progress({','.join(columns)}) VALUES ({placeholders})",
+                    tuple(payload[name] for name in columns),
+                )
+            else:
+                assignments = ", ".join(f"{name} = ?" for name in fields)
+                connection.execute(
+                    f"UPDATE eod_progress SET {assignments}, updated_at = ? WHERE trade_date = ?",
+                    (*fields.values(), isoformat(), day),
+                )
+
+    # ------------------------------------------------------------------
+    # Job execution
+    # ------------------------------------------------------------------
+
     def run_job(self, job_id: str, trade_date: str) -> dict[str, Any]:
         fn = self.jobs.get(job_id)
         if fn is None:
             raise KeyError(job_id)
+        if job_id in EOD_JOB_IDS:
+            if not self._eod_lock.acquire(blocking=False):
+                return {
+                    "job_id": job_id,
+                    "state": "skipped",
+                    "detail": "another EOD job is already running",
+                }
+            try:
+                return self._run_job(fn, job_id, trade_date)
+            finally:
+                self._eod_lock.release()
+        return self._run_job(fn, job_id, trade_date)
+
+    def _run_job(self, fn: JobFn, job_id: str, trade_date: str) -> dict[str, Any]:
         started = isoformat()
         run_id = None
         with self.db.write() as connection:
@@ -103,6 +287,10 @@ class Scheduler:
             run_id = cur.lastrowid
         try:
             result = fn(trade_date) or {}
+            if isinstance(result, dict):
+                failures = self.pipeline.eod_failures(result) if job_id in EOD_JOB_IDS else []
+                if failures:
+                    raise RuntimeError("; ".join(failures))
             with self.db.write() as connection:
                 connection.execute(
                     "UPDATE job_runs SET state=?, finished_at=?, rows_out=?, detail=? WHERE id=?",
@@ -126,6 +314,9 @@ class Scheduler:
     def _eod_b(self, trade_date: str) -> dict[str, Any]:
         return self.pipeline.run_eod_batch_b(trade_date)
 
+    def _eod_retry(self, trade_date: str) -> dict[str, Any]:
+        return self.pipeline.run_eod_missing(trade_date)
+
     def _history_backfill(self, trade_date: str) -> dict[str, Any]:
         return self.pipeline.backfill_history(trade_date)