baodan/api/insurance/poster/tasks.py

"""海报后台任务管理。"""
import json
import logging
import os
from datetime import datetime

from insurance.db.compat import db

logger = logging.getLogger(__name__)

# 超过此时间仍为 queued/generating 的任务视为过期（秒）
STALE_TASK_TIMEOUT = 600  # 10 分钟


def recover_stale_tasks():
    """启动时恢复过期任务：将长时间 queued/generating 的任务标记为 failed。

    应在应用启动时调用。
    """
    from insurance.models.poster_record import PosterRecord
    from insurance.models.poster_case_upload import PosterCaseUpload
    from sqlalchemy import inspect

    cutoff = datetime.now().timestamp() - STALE_TASK_TIMEOUT

    # 恢复过期的海报生成任务
    stale_records = PosterRecord.query.filter(
        PosterRecord.task_status.in_(["queued", "generating"]),
        PosterRecord.created_at < datetime.fromtimestamp(cutoff),
    ).all()
    for record in stale_records:
        record.task_status = "failed"
        record.task_error = "任务因服务重启而中断，请重新生成"
        record.finished_at = datetime.now()
        logger.warning(f"恢复过期海报任务: record_id={record.id}")

    # 恢复过期的计划书解析任务
    case_columns = {
        item["name"] for item in inspect(db.engine).get_columns("poster_case_uploads")
    }
    stale_cases = []
    if {"parse_progress", "parse_message", "parse_error"}.issubset(case_columns):
        stale_cases = PosterCaseUpload.query.filter(
            PosterCaseUpload.parse_status.in_(["queued", "parsing"]),
            PosterCaseUpload.created_at < datetime.fromtimestamp(cutoff),
        ).all()
    for case in stale_cases:
        case.parse_status = "failed"
        case.parse_progress = 100
        case.parse_message = "解析任务已中断"
        case.parse_error = "任务因服务重启或 Worker 中断，请重新解析"
        case.parse_finished_at = datetime.now()
        logger.warning(f"恢复过期解析任务: case_id={case.id}")

    if stale_records or stale_cases:
        db.session.commit()
        logger.info(f"已恢复 {len(stale_records)} 个海报任务, {len(stale_cases)} 个解析任务")


# ─── 计划书解析任务 ───────────────────────────────────────

def start_case_parse_task(case_upload_id: int) -> bool:
    """将计划书解析提交到 Celery Worker。"""
    try:
        from insurance.generation.celery_tasks import parse_poster_case_task
        from insurance.models.poster_case_upload import PosterCaseUpload

        result = parse_poster_case_task.apply_async(
            args=[case_upload_id],
            queue="insurance",
        )
        record = PosterCaseUpload.query.get(case_upload_id)
        if record:
            record.parse_task_id = result.id
            db.session.commit()
        return True
    except Exception as exc:
        logger.error("海报计划书解析任务提交失败 [%s]: %s", case_upload_id, exc, exc_info=True)
        return False


def _execute_case_parse(case_upload_id: int):
    from insurance.models.poster_case_upload import PosterCaseUpload

    record = PosterCaseUpload.query.get(case_upload_id)
    if not record:
        return

    filepath = record.source_file_url
    if not filepath or not os.path.exists(filepath):
        raise FileNotFoundError("计划书源文件不存在")

    from insurance.ppt.extraction import ExtractionOrchestrator
    orchestrator = ExtractionOrchestrator(use_cache=False)

    import asyncio
    parsed = None

    product_context = {}
    if record.product_snapshot_json:
        try:
            product_context = json.loads(record.product_snapshot_json) or {}
        except (json.JSONDecodeError, TypeError):
            logger.warning("计划书产品快照无效: case_id=%s", case_upload_id)
    plan_type = str(product_context.get("planType") or "savings").lower()
    product_data = product_context.get("productData") or {}
    product_name = (
        product_context.get("productName")
        or product_data.get("displayName")
        or ""
    )
    product_aliases = product_data.get("aliases") or []
    company_id = (
        product_context.get("companyId")
        or product_data.get("companyId")
        or ""
    )

    def report_progress(progress: int, message: str):
        _update_case_progress(case_upload_id, progress, message)

    # 优先使用海报紧凑解析，只提取基础字段和代表性利益年度。
    try:
        compact_data = asyncio.run(orchestrator.extract_for_poster(
            filepath,
            plan_type=plan_type,
            progress_callback=report_progress,
        ))
        compact_data = compact_data or {}
        if product_name and not compact_data.get("product_name"):
            compact_data["product_name"] = product_name
        parsed = _map_extract_plan_fields(
            compact_data, plan_type, "poster_compact", product_context=product_context,
        )
        if (parsed.get("meta") or {}).get("status") == "failed":
            parsed = None
        else:
            logger.info("使用海报紧凑解析成功: case_id=%s, type=%s", case_upload_id, plan_type)
    except Exception as exc:
        logger.warning("海报紧凑解析失败，降级到完整解析: %s", exc)

    # 降级：完整解析仍跳过海报不使用的提领方案和销售分析。
    if not parsed:
        try:
            result = asyncio.run(orchestrator.extract_plan(
                filepath,
                plan_type=plan_type,
                progress_callback=report_progress,
                company_id=company_id,
                product_id=record.product_id or "",
                product_name_hint=product_name,
                product_aliases=product_aliases,
                include_withdrawal=False,
            ))
            if result.status != "error" and result.data:
                parsed = _map_extract_plan_fields(
                    result.data, result.plan_type, result.status, product_context=product_context,
                )
                logger.info("使用完整解析降级成功: case_id=%s, type=%s", case_upload_id, result.plan_type)
        except Exception as exc:
            logger.error("完整解析降级也失败: %s", exc)

    if not parsed:
        raise ValueError("计划书中未识别到可用于海报的字段")

    record = PosterCaseUpload.query.get(case_upload_id)
    if not record:
        return
    parse_status = (parsed.get("meta") or {}).get("status", "failed")
    record.parsed_data = json.dumps(parsed, ensure_ascii=False)
    record.parse_status = parse_status
    record.parse_progress = 100
    record.parse_message = "解析完成" if parse_status == "parsed" else "解析完成，部分字段需核对"
    record.parse_error = None
    record.parse_heartbeat_at = datetime.now()
    record.parse_finished_at = datetime.now()
    db.session.commit()


def _map_extract_plan_fields(
    data: dict,
    plan_type: str,
    status: str,
    product_context: dict | None = None,
) -> dict:
    """将 extract_plan 的完整数据映射为海报前端所需字段结构。

    兼容多种可能的字段命名（LLM 输出不固定）。
    """
    insured = data.get("insured") or {}
    policy = data.get("policy") or {}
    product_context = product_context or {}
    product_data = product_context.get("productData") or {}
    company_data = product_context.get("companyData") or {}

    def first_value(*values):
        for value in values:
            if value is not None and value != "":
                return value
        return None

    raw_gender = first_value(insured.get("gender"), data.get("gender"))
    gender_map = {
        "male": "男", "m": "男", "男": "男",
        "female": "女", "f": "女", "女": "女",
    }
    gender = gender_map.get(str(raw_gender).strip().lower(), raw_gender) if raw_gender else None

    raw_currency = first_value(policy.get("currency"), data.get("currency"))
    from insurance.ppt.normalizer import _normalize_currency
    currency = _normalize_currency(raw_currency)

    # 保额：优先读取正式提取结构中的 policy
    sum_assured = (
        policy.get("sum_insured")
        or policy.get("basic_sum_insured")
        or policy.get("sum_assured")
        or data.get("sum_insured")
        or data.get("basic_sum_insured")
        or data.get("sum_assured")
        or data.get("face_amount")
        or data.get("coverage_amount")
    )
    # 缴费年期
    premium_term = (
        policy.get("premium_payment_period")
        or policy.get("premium_term")
        or data.get("premium_term")
        or data.get("payment_period")
        or data.get("paying_period")
        or data.get("premium_payment_term")
    )
    # 年缴保费
    annual_premium = (
        policy.get("annual_premium")
        or policy.get("basic_plan_annual_premium")
        or policy.get("target_premium")
        or data.get("annual_premium")
        or data.get("premium_amount")
        or data.get("yearly_premium")
    )
    # 保障期限
    coverage_period = (
        policy.get("coverage_period")
        or data.get("coverage_period")
        or data.get("policy_term")
        or data.get("coverage_term")
    )

    # 利益演示表
    benefit_table = (
        data.get("benefit_illustration")
        or data.get("benefit_table")
        or data.get("cash_value_projection")
        or []
    )

    smoking_status = first_value(
        insured.get("smoking_status"), insured.get("smoker_status"),
        insured.get("smoker"), insured.get("is_smoker"),
        data.get("smoking_status"), data.get("is_smoker"),
    )
    if isinstance(smoking_status, bool):
        smoking_status = "吸烟" if smoking_status else "非吸烟"
    elif smoking_status is not None:
        smoking_status = {
            "yes": "吸烟", "y": "吸烟", "smoker": "吸烟", "吸烟": "吸烟", "吸煙": "吸烟",
            "no": "非吸烟", "n": "非吸烟", "non-smoker": "非吸烟",
            "不吸烟": "非吸烟", "不吸煙": "非吸烟", "非吸烟": "非吸烟",
        }.get(str(smoking_status).strip().lower(), smoking_status)

    explicit_total_premium = first_value(
        policy.get("total_premium"), policy.get("total_basic_premium"),
        data.get("total_premium"), data.get("total_basic_premium"),
    )
    total_premium = explicit_total_premium
    if total_premium is None and isinstance(annual_premium, (int, float)) and isinstance(premium_term, (int, float)):
        total_premium = annual_premium * premium_term

    def benefit_value(year: int, *keys: str):
        for row in benefit_table:
            row_year = first_value(row.get("policy_year"), row.get("policyYear"), row.get("year"))
            try:
                if int(row_year) != year:
                    continue
            except (TypeError, ValueError):
                continue
            return first_value(*(row.get(key) for key in keys))
        return None

    explicit_initial_death_benefit = first_value(
        policy.get("initial_death_benefit"), data.get("initial_death_benefit"),
    )
    initial_death_benefit = first_value(
        explicit_initial_death_benefit,
        benefit_value(1, "death_benefit", "deathBenefit", "non_guaranteed_death_benefit"),
    )
    milestone_values = {
        year: first_value(
            data.get(f"surrender_value_{year}"),
            benefit_value(year, "total_surrender_value", "totalSurrenderValue", "total_surrender"),
        )
        for year in (10, 20, 30)
    }

    result = {
        "age": first_value(insured.get("age"), data.get("age")),
        "gender": gender,
        "smoking_status": smoking_status,
        "currency": currency,
        "company_name": first_value(
            product_context.get("companyName"), company_data.get("displayName"), data.get("company_name"),
        ),
        "product_name": first_value(
            product_context.get("productName"), product_data.get("displayName"), data.get("product_name"),
        ),
        "sum_assured": sum_assured,
        "basic_sum_assured": first_value(policy.get("basic_sum_insured"), data.get("basic_sum_insured"), sum_assured),
        "premium_term": premium_term,
        "annual_premium": annual_premium,
        "basic_plan_annual_premium": first_value(
            policy.get("basic_plan_annual_premium"), data.get("basic_plan_annual_premium"), annual_premium,
        ),
        "total_premium": total_premium,
        "first_year_amount_due": first_value(
            policy.get("first_year_amount_due"), data.get("first_year_amount_due"),
        ),
        "coverage_period": coverage_period,
        "initial_death_benefit": initial_death_benefit,
        "surrender_value_10": milestone_values[10],
        "surrender_value_20": milestone_values[20],
        "surrender_value_30": milestone_values[30],
        "key_benefits": data.get("key_benefits") or [],
        "benefit_table": benefit_table,
        "plan_type": plan_type,
        "extraction_status": status,
    }
    business_fields = (
        "age", "gender", "currency", "sum_assured",
        "premium_term", "annual_premium", "coverage_period",
    )
    valid_count = sum(
        1 for key in business_fields
        if result.get(key) is not None and result.get(key) != ""
    )
    required_by_type = {
        "savings": ("age", "currency", "annual_premium", "premium_term"),
        "ci": ("age", "currency", "sum_assured"),
        "iul": ("age", "currency", "sum_assured", "annual_premium"),
    }
    required = required_by_type.get(str(plan_type or "").lower(), ("age", "currency"))
    missing = [
        key for key in required
        if result.get(key) is None or result.get(key) == ""
    ]
    mapped_status = "failed" if valid_count == 0 else ("partial" if missing else "parsed")
    raw_provenance = data.get("_provenance") or data.get("provenance") or {}
    provenance = dict(raw_provenance) if isinstance(raw_provenance, dict) else {}
    if total_premium is not None and explicit_total_premium is None:
        provenance["total_premium"] = {
            "source": "system_derived",
            "formula": "annual_premium * premium_term",
        }
    if initial_death_benefit is not None and explicit_initial_death_benefit is None:
        provenance["initial_death_benefit"] = {
            "source": "system_derived",
            "from": "benefit_table.year_1",
        }
    for year, value in milestone_values.items():
        if value is not None and data.get(f"surrender_value_{year}") is None:
            provenance[f"surrender_value_{year}"] = {
                "source": "system_derived",
                "from": f"benefit_table.year_{year}",
            }

    result["meta"] = {
        "planType": plan_type,
        "status": mapped_status,
        "sourceStatus": status,
        "missingFields": missing,
        "validFieldCount": valid_count,
        "method": (
            (data.get("_meta") or {}).get("method")
            or (data.get("extraction_meta") or {}).get("method")
            or status
        ),
        "provenance": provenance,
        "lowQualityPages": (
            (data.get("_meta") or {}).get("low_quality_pages")
            or (data.get("extraction_meta") or {}).get("lowQualityPages")
            or []
        ),
    }
    return result


def _mark_case_failed(case_upload_id: int, error: str):
    from insurance.models.poster_case_upload import PosterCaseUpload

    record = PosterCaseUpload.query.get(case_upload_id)
    if not record:
        return
    record.parse_status = "failed"
    record.parse_progress = 100
    record.parse_message = "解析失败"
    record.parse_error = error[:1000]
    record.parse_heartbeat_at = datetime.now()
    record.parse_finished_at = datetime.now()
    db.session.commit()


def _update_case_progress(case_upload_id: int, progress: int, message: str):
    from insurance.models.poster_case_upload import PosterCaseUpload

    record = PosterCaseUpload.query.get(case_upload_id)
    if not record:
        return
    record.parse_progress = max(0, min(int(progress), 99))
    record.parse_message = message[:500]
    record.parse_heartbeat_at = datetime.now()
    db.session.commit()