zhixing-system/zhixing-server/src/zhixing_server/modules/selection/application/run.py

"""Application orchestration for persisted whole-universe strategy runs."""

from __future__ import annotations

import logging
import time
from collections import Counter
from collections.abc import Callable, Mapping, Sequence
from concurrent.futures import ThreadPoolExecutor
from dataclasses import dataclass, replace
from datetime import date
from typing import Protocol, cast

from ..domain.models import SelectionEvaluation, SelectionStrategyName, StockHistory
from ..domain.pattern_scoring import (
    PatternCase,
    PatternCaseLibraryLoader,
    PatternScore,
    PatternScorer,
)
from ..domain.runs import (
    BatchSelectionRunStore,
    BatchSelectionUniverseReader,
    SelectionExecutionSource,
    SelectionRerunRequired,
    SelectionResultQuery,
    SelectionRun,
    SelectionRunInProgress,
    SelectionRunItem,
    SelectionRunStatus,
    SelectionRunStore,
    SelectionSectorCount,
    SelectionSectorMembership,
    SelectionSectorReader,
    SelectionStock,
    SelectionUniverseReader,
)
from .evaluate import EvaluateZhixingB1

# Selection runs are started by the ASGI service in production.  A child of
# Uvicorn's configured logger keeps INFO diagnostics visible in container logs.
logger = logging.getLogger("uvicorn.error.zhixing.selection.run")
StrategyName = SelectionStrategyName
_FAILURE_STATUSES = {
    "insufficient_history",
    "missing_target_bar",
    "missing_turnover_rate",
    "data_error",
}


class SelectionEvaluator(Protocol):
    """Minimal single-stock evaluator required by the batch orchestrator."""

    def execute(self, ts_code: str, target_trade_date: date) -> SelectionEvaluation: ...


@dataclass(frozen=True, slots=True)
class PreparedSelectionRun:
    """A claimed run and its immutable market-data source snapshot."""

    run: SelectionRun
    source: SelectionExecutionSource


@dataclass(frozen=True, slots=True)
class SelectionSectorAggregates:
    """A run summary plus its selected stocks' sector membership counts."""

    run: SelectionRun | None
    snapshot_trade_date: date | None
    sector_type: str
    sectors: tuple[SelectionSectorCount, ...]


class RunZhixingB1:
    """Prepare, execute, and query persisted selection strategy batches.

    The historical class name remains as a compatibility seam for existing
    composition and tests while strategy routing is now explicit.
    """

    def __init__(
        self,
        reader: SelectionUniverseReader,
        store: SelectionRunStore,
        evaluator: SelectionEvaluator | None = None,
        evaluators: Mapping[StrategyName, SelectionEvaluator] | None = None,
        pattern_case_loader: PatternCaseLibraryLoader | None = None,
        pattern_scorer: PatternScorer | None = None,
        sector_reader: SelectionSectorReader | None = None,
        *,
        pattern_scoring_enabled: bool = False,
        max_workers: int = 4,
        batch_size: int = 200,
    ) -> None:
        """Inject storage ports and configure bounded chunk execution."""

        if max_workers < 1:
            raise ValueError("max_workers must be at least 1")
        if batch_size < 1:
            raise ValueError("batch_size must be at least 1")
        self.reader = reader
        self.store = store
        self.evaluator = evaluator or EvaluateZhixingB1(reader)
        self.evaluators: dict[StrategyName, SelectionEvaluator] = {
            "zhixing_b1": self.evaluator,
        }
        if evaluators is not None:
            self.evaluators.update(evaluators)
        self.pattern_case_loader = pattern_case_loader
        self.pattern_scorer = pattern_scorer
        self.sector_reader = sector_reader
        self.pattern_scoring_enabled = pattern_scoring_enabled
        self.max_workers = max_workers
        self.batch_size = batch_size

    def prepare(
        self,
        strategy: StrategyName,
        target_trade_date: date,
        *,
        rerun: bool,
    ) -> PreparedSelectionRun:
        """Validate source eligibility before claiming the rerunnable key."""

        logger.info(
            "selection_run_prepare_started strategy=%s target_trade_date=%s rerun=%s",
            strategy,
            target_trade_date.isoformat(),
            rerun,
        )
        if strategy not in self.evaluators:
            raise ValueError(f"selection evaluator is not configured for strategy: {strategy}")
        try:
            source = self.reader.load_execution_source(strategy, target_trade_date)
            run = self.store.prepare_run(
                strategy,
                target_trade_date,
                source,
                rerun=rerun,
            )
        except Exception as exc:  # noqa: BLE001 - log the safe prepare boundary and preserve type
            logger.warning(
                "selection_run_prepare_failed strategy=%s target_trade_date=%s "
                "status=failed error_type=%s reason=%s",
                strategy,
                target_trade_date.isoformat(),
                exc.__class__.__name__,
                _safe_item_error(exc),
            )
            raise
        logger.info(
            "selection_run_prepared strategy=%s target_trade_date=%s run_id=%s "
            "market_sync_batch_id=%s target_count=%d eligible_count=%d "
            "coverage=%s status=running",
            strategy,
            target_trade_date.isoformat(),
            run.id,
            source.market_sync_batch_id,
            source.target_count,
            len(source.stocks),
            source.coverage,
        )
        return PreparedSelectionRun(run=run, source=source)

    def execute(self, prepared: PreparedSelectionRun) -> None:
        """Evaluate every eligible stock and converge the persisted run status.

        This method is the boundary used by FastAPI's in-process background
        task.  An unexpected batch-level error is recorded before the worker
        returns so the UI never mistakes a lost worker exception for success.
        """

        stocks = _unique_stocks(prepared.source.stocks)
        strategy = prepared.run.strategy
        target_trade_date = prepared.source.target_trade_date
        evaluator = self.evaluators[strategy]
        evaluated_count = 0
        selected_stock_count = 0
        signal_count = 0
        failed_count = 0
        missing_turnover_count = 0
        insufficient_history_count = 0
        history_rows = 0
        batch_count = _chunk_count(len(stocks), self.batch_size)
        current_batch = 0
        final_status: SelectionRunStatus = "failed"
        read_seconds = 0.0
        evaluate_seconds = 0.0
        persist_seconds = 0.0
        scoring_seconds = 0.0
        logger.info(
            "selection_run_started strategy=%s target_trade_date=%s run_id=%s "
            "market_sync_batch_id=%s stock_count=%d batch_count=%d worker_count=%d "
            "status=running",
            strategy,
            target_trade_date.isoformat(),
            prepared.run.id,
            prepared.source.market_sync_batch_id,
            len(stocks),
            batch_count,
            self.max_workers,
        )
        try:
            if strategy == "zhixing_b1":
                pattern_cases, pattern_library_error = self._prepare_pattern_cases(prepared.run.id)
            else:
                pattern_cases, pattern_library_error = None, None
            with ThreadPoolExecutor(max_workers=self.max_workers) as executor:
                for batch_index, batch_stocks in enumerate(
                    _chunks(stocks, self.batch_size),
                    start=1,
                ):
                    current_batch = batch_index
                    read_started = time.perf_counter()
                    histories = self._load_histories(
                        batch_stocks,
                        target_trade_date,
                        evaluator,
                    )
                    batch_read_seconds = time.perf_counter() - read_started
                    read_seconds += batch_read_seconds
                    batch_history_rows = sum(
                        len(history.bars) for history in histories if history is not None
                    )
                    history_rows += batch_history_rows
                    batch_missing_turnover = sum(
                        not _turnover_present(history, target_trade_date) for history in histories
                    )
                    logger.info(
                        "selection_read_batch_summary strategy=%s target_trade_date=%s "
                        "run_id=%s batch=%d batch_count=%d stock_count=%d history_rows=%d "
                        "turnover_missing_count=%d status=success read_seconds=%.3f",
                        strategy,
                        target_trade_date.isoformat(),
                        prepared.run.id,
                        batch_index,
                        batch_count,
                        len(batch_stocks),
                        batch_history_rows,
                        batch_missing_turnover,
                        batch_read_seconds,
                    )

                    evaluate_started = time.perf_counter()
                    evaluations = tuple(
                        executor.map(
                            self._evaluate_stock,
                            batch_stocks,
                            histories,
                            [target_trade_date] * len(batch_stocks),
                            [evaluator] * len(batch_stocks),
                            [strategy] * len(batch_stocks),
                            [prepared.run.id] * len(batch_stocks),
                            [batch_index] * len(batch_stocks),
                        )
                    )
                    batch_evaluate_seconds = time.perf_counter() - evaluate_started
                    evaluate_seconds += batch_evaluate_seconds
                    status_counts = Counter(evaluation.status for evaluation in evaluations)
                    no_signal_reasons = Counter(
                        evaluation.reason or "unspecified"
                        for evaluation in evaluations
                        if evaluation.status == "no_signal"
                    )
                    missing_turnover_count += status_counts["missing_turnover_rate"]
                    insufficient_history_count += status_counts["insufficient_history"]
                    for stock, history, evaluation in zip(
                        batch_stocks,
                        histories,
                        evaluations,
                        strict=True,
                    ):
                        if evaluation.status not in _FAILURE_STATUSES:
                            continue
                        logger.warning(
                            "selection_item_incomplete strategy=%s target_trade_date=%s "
                            "run_id=%s batch=%d ts_code=%s history_rows=%d "
                            "turnover_present=%s status=%s error_type=%s reason=%s",
                            strategy,
                            target_trade_date.isoformat(),
                            prepared.run.id,
                            batch_index,
                            stock.ts_code,
                            len(history.bars) if history is not None else 0,
                            _turnover_present(history, target_trade_date),
                            evaluation.status,
                            evaluation.status,
                            evaluation.reason or evaluation.status,
                        )

                    scoring_started = time.perf_counter()
                    items = tuple(
                        _to_item(
                            stock.ts_code,
                            stock.name,
                            evaluation,
                            pattern_score=self._score_stock(
                                strategy,
                                prepared.run.id,
                                stock,
                                history,
                                evaluation,
                                pattern_cases,
                                pattern_library_error,
                            ),
                        )
                        for stock, history, evaluation in zip(
                            batch_stocks,
                            histories,
                            evaluations,
                            strict=True,
                        )
                    )
                    scoring_seconds += time.perf_counter() - scoring_started

                    evaluated_count += len(items)
                    selected_stock_count += sum(item.status == "selected" for item in items)
                    signal_count += sum(item.signal_count for item in items)
                    failed_count += sum(item.status in _FAILURE_STATUSES for item in items)

                    logger.info(
                        "selection_evaluate_batch_summary strategy=%s target_trade_date=%s "
                        "run_id=%s batch=%d batch_count=%d stock_count=%d selected_count=%d "
                        "no_signal_count=%d insufficient_history_count=%d "
                        "missing_target_bar_count=%d missing_turnover_count=%d "
                        "data_error_count=%d no_signal_reasons=%s status=complete "
                        "evaluate_seconds=%.3f",
                        strategy,
                        target_trade_date.isoformat(),
                        prepared.run.id,
                        batch_index,
                        batch_count,
                        len(items),
                        status_counts["selected"],
                        status_counts["no_signal"],
                        status_counts["insufficient_history"],
                        status_counts["missing_target_bar"],
                        status_counts["missing_turnover_rate"],
                        status_counts["data_error"],
                        dict(no_signal_reasons),
                        batch_evaluate_seconds,
                    )

                    persist_started = time.perf_counter()
                    self._record_items(prepared.run.id, items)
                    batch_persist_seconds = time.perf_counter() - persist_started
                    persist_seconds += batch_persist_seconds
                    logger.info(
                        "selection_persist_batch_summary strategy=%s target_trade_date=%s "
                        "run_id=%s batch=%d batch_count=%d item_count=%d signal_count=%d "
                        "status=success persist_seconds=%.3f",
                        strategy,
                        target_trade_date.isoformat(),
                        prepared.run.id,
                        batch_index,
                        batch_count,
                        len(items),
                        sum(item.signal_count for item in items),
                        batch_persist_seconds,
                    )

            final_status = _run_status(evaluated_count, failed_count)
            self.store.finish_run(
                prepared.run.id,
                final_status,
                evaluated_count=evaluated_count,
                selected_stock_count=selected_stock_count,
                signal_count=signal_count,
                failed_count=failed_count,
            )
            logger.info(
                "selection_run_converged strategy=%s target_trade_date=%s run_id=%s "
                "batch=%d evaluated_count=%d selected_stock_count=%d signal_count=%d "
                "failed_count=%d status=%s",
                strategy,
                target_trade_date.isoformat(),
                prepared.run.id,
                current_batch,
                evaluated_count,
                selected_stock_count,
                signal_count,
                failed_count,
                final_status,
            )
        except Exception as exc:  # noqa: BLE001 - worker boundary must persist failure state
            final_status = "failed"
            logger.error(
                "selection_run_failed strategy=%s target_trade_date=%s run_id=%s "
                "batch=%d status=failed error_type=%s reason=%s",
                strategy,
                target_trade_date.isoformat(),
                prepared.run.id,
                current_batch,
                exc.__class__.__name__,
                _safe_item_error(exc),
            )
            try:
                self.store.finish_run(
                    prepared.run.id,
                    "failed",
                    evaluated_count=evaluated_count,
                    selected_stock_count=selected_stock_count,
                    signal_count=signal_count,
                    failed_count=max(failed_count, 1),
                    error_type="batch_error",
                    error_message=_safe_item_error(exc),
                )
            except Exception:  # noqa: BLE001 - preserve the original worker failure
                logger.error(
                    "selection_run_failure_persist_failed strategy=%s target_trade_date=%s "
                    "run_id=%s batch=%d status=failed error_type=finish_run_failed",
                    strategy,
                    target_trade_date.isoformat(),
                    prepared.run.id,
                    current_batch,
                )
        finally:
            logger.info(
                "selection_run_summary strategy=%s target_trade_date=%s run_id=%s "
                "market_sync_batch_id=%s stock_count=%d history_rows=%d batch_count=%d "
                "last_batch=%d worker_count=%d evaluated_count=%d selected_stock_count=%d "
                "signal_count=%d failed_count=%d insufficient_history_count=%d "
                "missing_turnover_count=%d status=%s read_seconds=%.3f "
                "evaluate_seconds=%.3f scoring_seconds=%.3f persist_seconds=%.3f",
                strategy,
                target_trade_date.isoformat(),
                prepared.run.id,
                prepared.source.market_sync_batch_id,
                len(stocks),
                history_rows,
                batch_count,
                current_batch,
                self.max_workers,
                evaluated_count,
                selected_stock_count,
                signal_count,
                failed_count,
                insufficient_history_count,
                missing_turnover_count,
                final_status,
                read_seconds,
                evaluate_seconds,
                scoring_seconds,
                persist_seconds,
            )

    def _prepare_pattern_cases(
        self,
        run_id: str,
    ) -> tuple[tuple[PatternCase, ...] | None, str | None]:
        """Load the complete case library once without failing selection."""

        if not self.pattern_scoring_enabled:
            return None, None
        if self.pattern_case_loader is None or self.pattern_scorer is None:
            reason = "pattern scoring is enabled but not configured"
            logger.error("selection_pattern_library_failed run_id=%s reason=%s", run_id, reason)
            return None, reason
        try:
            return self.pattern_case_loader.load(), None
        except Exception as exc:  # noqa: BLE001 - scoring enrichment must not fail selection
            reason = _safe_item_error(exc)
            logger.warning(
                "selection_pattern_library_failed run_id=%s error_type=%s reason=%s",
                run_id,
                exc.__class__.__name__,
                reason,
            )
            return None, reason

    def _score_stock(
        self,
        strategy: StrategyName,
        run_id: str,
        stock: SelectionStock,
        history: StockHistory | None,
        evaluation: SelectionEvaluation,
        cases: tuple[PatternCase, ...] | None,
        library_error: str | None,
    ) -> PatternScore:
        """Score one selected stock once and isolate enrichment failures."""

        if (
            strategy != "zhixing_b1"
            or not self.pattern_scoring_enabled
            or evaluation.status != "selected"
        ):
            return PatternScore()
        if library_error is not None:
            return PatternScore.failed(library_error)
        if history is None or cases is None or self.pattern_scorer is None:
            return PatternScore.failed("pattern scoring history or case library is unavailable")
        try:
            return self.pattern_scorer.score(history, cases)
        except Exception as exc:  # noqa: BLE001 - one score must not fail the selection run
            reason = _safe_item_error(exc)
            logger.warning(
                "selection_pattern_score_failed run_id=%s ts_code=%s error_type=%s reason=%s",
                run_id,
                stock.ts_code,
                exc.__class__.__name__,
                reason,
            )
            return PatternScore.failed(reason)

    def _load_histories(
        self,
        stocks: Sequence[SelectionStock],
        target_trade_date: date,
        evaluator: SelectionEvaluator,
    ) -> tuple[StockHistory | None, ...]:
        """Load one chunk when the reader supports it, with old-path fallback."""

        typed_stocks = tuple(stocks)
        loader = getattr(self.reader, "load_histories", None)
        if callable(loader):
            batch_reader = cast(BatchSelectionUniverseReader, self.reader)
            loaded = batch_reader.load_histories(typed_stocks, target_trade_date)
            histories_by_code = {history.ts_code: history for history in loaded}
            return tuple(
                histories_by_code.get(
                    stock.ts_code,
                    StockHistory(ts_code=stock.ts_code, name=stock.name),
                )
                for stock in typed_stocks
            )

        execute_history = getattr(evaluator, "execute_history", None)
        if callable(execute_history):
            return tuple(
                self.reader.load_history(stock.ts_code, target_trade_date) for stock in typed_stocks
            )
        return (None,) * len(typed_stocks)

    def _evaluate_stock(
        self,
        stock: SelectionStock,
        history: StockHistory | None,
        target_trade_date: date,
        evaluator: SelectionEvaluator,
        strategy: StrategyName,
        run_id: str,
        batch_index: int,
    ) -> SelectionEvaluation:
        """Evaluate one stock inside a worker and isolate its exception."""

        ts_code = stock.ts_code
        try:
            execute_history: Callable[[StockHistory, date], SelectionEvaluation] | None = getattr(
                evaluator,
                "execute_history",
                None,
            )
            if history is not None and execute_history is not None:
                return execute_history(history, target_trade_date)
            return evaluator.execute(ts_code, target_trade_date)
        except Exception as exc:  # noqa: BLE001 - isolate one stock from the batch
            logger.warning(
                "selection_item_failed strategy=%s target_trade_date=%s run_id=%s "
                "batch=%d ts_code=%s history_rows=%d turnover_present=%s "
                "status=data_error error_type=%s reason=%s",
                strategy,
                target_trade_date.isoformat(),
                run_id,
                batch_index,
                ts_code,
                len(history.bars) if history is not None else 0,
                _turnover_present(history, target_trade_date),
                exc.__class__.__name__,
                _safe_item_error(exc),
            )
            return SelectionEvaluation(
                ts_code=ts_code,
                target_trade_date=target_trade_date,
                status="data_error",
                reason=_safe_item_error(exc),
            )

    def _record_items(self, run_id: str, items: Sequence[SelectionRunItem]) -> None:
        """Use batch persistence while retaining the old single-item seam."""

        record_items = getattr(self.store, "record_items", None)
        if callable(record_items):
            batch_store = cast(BatchSelectionRunStore, self.store)
            batch_store.record_items(run_id, tuple(items))
            return
        for item in items:
            self.store.record_item(run_id, item)

    def get_run(
        self,
        run_id: str,
        *,
        query: SelectionResultQuery | None = None,
    ) -> SelectionRun | None:
        """Read one persisted run for polling with optional sector filtering."""

        effective = query or SelectionResultQuery()
        if not effective.sector:
            return self.store.get_run(run_id, query=query)
        identity = self.store.get_run_identity(run_id)
        if identity is None:
            return None
        member_codes = self._sector_member_codes(identity.target_trade_date, effective.sector)
        if member_codes is None:
            return self.store.get_run(run_id, query=query)
        return self.store.get_run(
            run_id,
            query=replace(effective, sector=None),
            sector_stock_codes=member_codes,
        )

    def get_latest(
        self,
        strategy: StrategyName,
        target_trade_date: date | None = None,
        *,
        query: SelectionResultQuery | None = None,
    ) -> SelectionRun | None:
        """Read the current result by date or the latest result for a strategy."""

        effective = query or SelectionResultQuery()
        if not effective.sector:
            return self.store.get_latest_run(strategy, target_trade_date, query=query)
        identity = self.store.get_latest_run_identity(strategy, target_trade_date)
        if identity is None:
            return None
        member_codes = self._sector_member_codes(identity.target_trade_date, effective.sector)
        if member_codes is None:
            return self.store.get_latest_run(strategy, target_trade_date, query=query)
        return self.store.get_latest_run(
            strategy,
            identity.target_trade_date,
            query=replace(effective, sector=None),
            sector_stock_codes=member_codes,
        )

    def _sector_member_codes(
        self,
        target_trade_date: date,
        sector_code: str,
    ) -> tuple[str, ...] | None:
        """Resolve one sector's members, or None when the port is absent."""

        if self.sector_reader is None:
            return None
        return self.sector_reader.sector_member_codes(target_trade_date, sector_code)

    def list_sector_counts(
        self,
        strategy: StrategyName,
        target_trade_date: date | None = None,
        *,
        sector_type: str = "concept",
    ) -> SelectionSectorAggregates | None:
        """Aggregate the current run's selected stocks by point-in-time sector."""

        run = self.store.get_latest_run(strategy, target_trade_date)
        if run is None:
            return None
        selected_codes = [
            item.ts_code
            for item in run.items
            if item.status == "selected" and item.signal_count > 0
        ]
        membership = self._sector_membership(selected_codes, run.target_trade_date, sector_type)
        return SelectionSectorAggregates(
            run=run,
            snapshot_trade_date=membership.snapshot_trade_date,
            sector_type=sector_type,
            sectors=membership.sector_counts,
        )

    def _sector_membership(
        self,
        stock_codes: Sequence[str],
        target_trade_date: date,
        sector_type: str,
    ) -> SelectionSectorMembership:
        """Read sector counts for a stock set, tolerating a missing port."""

        if self.sector_reader is None or not stock_codes:
            return SelectionSectorMembership(snapshot_trade_date=None, sector_counts=())
        return self.sector_reader.sector_counts(
            stock_codes,
            target_trade_date,
            sector_type=sector_type,
        )


def _to_item(
    ts_code: str,
    name: str,
    evaluation: SelectionEvaluation,
    *,
    pattern_score: PatternScore | None = None,
) -> SelectionRunItem:
    """Translate a single-stock domain result into a stored item."""

    return SelectionRunItem(
        ts_code=ts_code,
        name=name or (evaluation.signals[0].name if evaluation.signals else ""),
        status=evaluation.status,
        signal_count=len(evaluation.signals),
        reason=evaluation.reason,
        pattern_score=pattern_score or PatternScore(),
        signals=evaluation.signals,
    )


def _run_status(evaluated_count: int, failed_count: int) -> SelectionRunStatus:
    """Map per-stock outcomes into a visible batch status."""

    if failed_count == 0:
        return "success"
    if evaluated_count == 0 or failed_count >= evaluated_count:
        return "failed"
    return "partial_success"


def _safe_item_error(error: Exception) -> str:
    """Keep per-stock failure context readable without persisting tracebacks."""

    return " ".join(str(error).split())[:500] or error.__class__.__name__


def _turnover_present(history: StockHistory | None, target_trade_date: date) -> bool:
    """Return whether target-day Tushare turnover is available for diagnostics."""

    if history is None:
        return False
    basic = history.daily_basic.get(target_trade_date)
    return basic is not None and basic.turnover_rate is not None


def _chunks(
    values: Sequence[SelectionStock],
    size: int,
) -> tuple[tuple[SelectionStock, ...], ...]:
    """Split a stable stock sequence into bounded immutable chunks."""

    return tuple(tuple(values[index : index + size]) for index in range(0, len(values), size))


def _chunk_count(value_count: int, size: int) -> int:
    """Return the number of chunks without materializing empty chunks."""

    return (value_count + size - 1) // size


def _unique_stocks(stocks: Sequence[SelectionStock]) -> tuple[SelectionStock, ...]:
    """Keep the first source row for each stock so it is evaluated once."""

    seen: set[str] = set()
    unique: list[SelectionStock] = []
    for stock in stocks:
        ts_code = stock.ts_code
        if ts_code in seen:
            continue
        seen.add(ts_code)
        unique.append(stock)
    return tuple(unique)


__all__ = [
    "PreparedSelectionRun",
    "RunZhixingB1",
    "SelectionRerunRequired",
    "SelectionRunInProgress",
    "SelectionSectorAggregates",
]