demand_server/app/services/element_search_service.py

"""实质元素需求检索：通过 ODPS（MaxCompute）执行 Hive SQL。"""

import json
import re
from datetime import date, datetime, timedelta
from decimal import Decimal
from zoneinfo import ZoneInfo

from app.odps.client import get_odps_client

_DATE_PARTITION_RE = re.compile(r"^\d{8}$")
SHANGHAI_TZ = ZoneInfo("Asia/Shanghai")


def _today_shanghai() -> date:
    return datetime.now(SHANGHAI_TZ).date()


def _same_day_last_year(value: date) -> date:
    try:
        return value.replace(year=value.year - 1)
    except ValueError:
        return date(value.year - 1, 2, 28)


def _partition_yyyymmdd(value: date) -> str:
    return value.strftime("%Y%m%d")


def _validate_partition_dt(label: str, value: str) -> str:
    if not _DATE_PARTITION_RE.match(value):
        raise ValueError(f"{label} 须为 YYYYMMDD 分区格式，当前为 {value!r}")
    return value


def _serialize_video_list(value: object) -> str | None:
    if value is None:
        return None
    if isinstance(value, list):
        return json.dumps(value, ensure_ascii=False)
    return str(value)


def _normalize_scalar(value: object) -> object:
    if isinstance(value, Decimal):
        return float(value)
    return value


def query_same_period_last_year_element_demands(
    *,
    period_days: int,
    view_pv_count: int,
    min_contribution_score: float | int,
    rov_avg: float | int,
) -> list[dict[str, object]]:
    """去年同期阳历策略下的实质元素需求聚合（ROV 等均值过滤）。

    分区 dt 范围（YYYYMMDD）由「上海时区下的去年今日」与时间段推算：
    - start_date：去年的今天（与今年同日历月日；闰年 2 月 29 日则退成去年 2 月 28 日）
    - end_date：从 start_date 起连续 period_days 个日历日的最后一天（含起始日）

    demand_id 中的 bizdate 固定为上海时区当天日期（YYYYMMDD），不作为入参。

    其余参数：
    - period_days：区间天数（含起始日），须 >= 0；为 0 仅含 start_date 当日；为 7 表示含 start_date 在内共 7 天分区
    - view_pv_count：`当日分发曝光pv` 下限
    - min_contribution_score：`贡献分` 下限
    - rov_avg：分组后平均 ROV 下限
    """
    if period_days < 0:
        raise ValueError("period_days 不能为负")

    anchor_last_year = _same_day_last_year(_today_shanghai())
    start_dt = anchor_last_year
    end_dt = anchor_last_year + timedelta(days=max(0, period_days - 1))

    start_date = _validate_partition_dt("start_date", _partition_yyyymmdd(start_dt))
    end_date = _validate_partition_dt("end_date", _partition_yyyymmdd(end_dt))
    bizdate = _partition_yyyymmdd(_today_shanghai())
    if view_pv_count < 0:
        raise ValueError("view_pv_count 不能为负")

    sql = f"""
WITH cleaned_video_metrics AS
(
    SELECT  CAST(视频id AS STRING) AS vid
            ,rov_t0
    FROM    loghubods.video_dimension_detail_add_column
    WHERE   dt >= '{start_date}'
    AND     dt <= '{end_date}'
    AND     COALESCE(`当日分发曝光pv`,0) >= {int(view_pv_count)}
)
,video_avg_metrics AS
(
    SELECT  vid
            ,AVG(CASE WHEN rov_t0 = 0 THEN NULL ELSE rov_t0 END) AS vid_avg_rov
    FROM    cleaned_video_metrics
    GROUP BY vid
)
,tag_vid_dedup AS
(
    SELECT  DISTINCT CAST(vid AS STRING) AS vid
            ,原始元素
            ,原始元素描述
            ,点类型
            ,元素维度
            ,短语
            ,选题
            ,`extend`
    FROM    loghubods.dwd_topic_decode_result_detail_di
    WHERE   dt = MAX_PT('loghubods.dwd_topic_decode_result_detail_di')
    AND     TRIM(原始元素) <> ''
    AND     原始元素 IS NOT NULL
    AND     元素维度 = '实质'
    AND     贡献分 >= {float(min_contribution_score)}
)
SELECT  '去年同期阳历' AS strategy
        ,md5(CONCAT('去年同期阳历',t1.原始元素,'{bizdate}')) AS demand_id
        ,t1.原始元素 AS demand_name
        ,COALESCE(ROUND(AVG(t2.vid_avg_rov),6),0) AS weight
        ,COUNT(DISTINCT t1.vid) AS video_count
        ,COLLECT_SET(t1.vid) AS video_list
        ,'{{}}' AS ext_info
FROM    tag_vid_dedup t1
LEFT JOIN video_avg_metrics t2
ON      t1.vid = t2.vid
GROUP BY t1.原始元素
HAVING  COALESCE(ROUND(AVG(t2.vid_avg_rov),6),0) >= {float(rov_avg)}
ORDER BY weight DESC
"""

    odps_client = get_odps_client()
    instance = odps_client.execute_sql(sql)

    rows: list[dict[str, object]] = []
    with instance.open_reader(tunnel=True) as reader:
        for record in reader:
            rows.append(
                {
                    "strategy": record["strategy"],
                    "demand_id": record["demand_id"],
                    "demand_name": record["demand_name"],
                    "weight": _normalize_scalar(record["weight"]),
                    "video_count": record["video_count"],
                    "video_list": _serialize_video_list(record["video_list"]),
                    "ext_info": record["ext_info"],
                }
            )
    return rows


def query_same_period_last_year_lunar_element_demands(
    *,
    period_days: int,
    view_pv_count: int,
    min_contribution_score: float | int,
    rov_avg: float | int,
) -> list[dict[str, object]]:
    """去年同期阴历策略下的实质元素需求聚合（ROV 等均值过滤）。

    日期区间由 SQL 内 lunar_to_solar / lunar_add / solar_to_lunar 推算；
    end_date 为 start_date 起连续 period_days 天的最后一天（含首日），与阳历接口语义一致。
    bizdate 为上海时区当天 YYYYMMDD。
    """
    if period_days < 0:
        raise ValueError("period_days 不能为负")
    if view_pv_count < 0:
        raise ValueError("view_pv_count 不能为负")

    bizdate = _validate_partition_dt("bizdate", _partition_yyyymmdd(_today_shanghai()))
    days_after = max(0, period_days - 1)

    sql = f"""
SET odps.sql.type.system.odps2 = true
;
SET odps.sql.decimal.odps2 = true
;

WITH base_date AS
(
    SELECT  lunar_to_solar(lunar_add(loghubods.solar_to_lunar('{bizdate}','yyyyMMdd'),-1,'Y')) AS start_date
)
,date_params AS
(
    SELECT  start_date
            ,DATE_FORMAT(CAST(DATE_ADD(TO_DATE(start_date,'yyyyMMdd'),{int(days_after)}) AS TIMESTAMP),'yyyyMMdd') AS end_date
    FROM    base_date
)
,cleaned_video_metrics AS
(
    SELECT  CAST(d.视频id AS STRING) AS vid
            ,d.rov_t0
    FROM    loghubods.video_dimension_detail_add_column d
    WHERE   d.dt >= (
                SELECT  start_date
                FROM    date_params
            )
    AND     d.dt <= (
                SELECT  end_date
                FROM    date_params
            )
    AND     COALESCE(d.`当日分发曝光pv`,0) >= {int(view_pv_count)}
)
,video_avg_metrics AS
(
    SELECT  vid
            ,AVG(CASE WHEN rov_t0 = 0 THEN NULL ELSE rov_t0 END) AS vid_avg_rov
    FROM    cleaned_video_metrics
    GROUP BY vid
)
,tag_vid_dedup AS
(
    SELECT  DISTINCT CAST(vid AS STRING) AS vid
            ,原始元素
            ,原始元素描述
            ,点类型
            ,元素维度
            ,短语
            ,选题
            ,`extend`
    FROM    loghubods.dwd_topic_decode_result_detail_di
    WHERE   dt = MAX_PT('loghubods.dwd_topic_decode_result_detail_di')
    AND     TRIM(原始元素) <> ''
    AND     原始元素 IS NOT NULL
    AND     元素维度 = '实质'
    AND     贡献分 >= {float(min_contribution_score)}
)
SELECT  '去年同期阴历' AS strategy
        ,md5(CONCAT('去年同期阴历',t1.原始元素,'{bizdate}')) AS demand_id
        ,t1.原始元素 AS demand_name
        ,CAST(COALESCE(ROUND(AVG(t2.vid_avg_rov), 6), 0) AS DECIMAL(20,6)) AS weight
        ,COUNT(DISTINCT t1.vid) AS video_count
        ,COLLECT_SET(t1.vid) AS video_list
        ,'{{}}' AS ext_info
FROM    tag_vid_dedup t1
LEFT JOIN video_avg_metrics t2
ON      t1.vid = t2.vid
GROUP BY t1.原始元素
HAVING  COALESCE(ROUND(AVG(t2.vid_avg_rov),6),0) >= {float(rov_avg)}
ORDER BY weight DESC
"""

    odps_client = get_odps_client()
    # 脚本模式：SET 多条 + SELECT；DECIMAL(p,s) 需 odps.sql.decimal.odps2
    instance = odps_client.execute_sql(
        sql,
        hints={
            "odps.sql.submit.mode": "script",
            "odps.sql.decimal.odps2": "true",
        },
    )

    rows: list[dict[str, object]] = []
    with instance.open_reader(tunnel=True) as reader:
        for record in reader:
            rows.append(
                {
                    "strategy": record["strategy"],
                    "demand_id": record["demand_id"],
                    "demand_name": record["demand_name"],
                    "weight": _normalize_scalar(record["weight"]),
                    "video_count": record["video_count"],
                    "video_list": _serialize_video_list(record["video_list"]),
                    "ext_info": record["ext_info"],
                }
            )
    return rows


def query_monthly_element_demands(
    *,
    view_pv_count: int,
    month_total_pv_threshold: float | int,
    min_contribution_score: float | int,
    rov_avg: float | int,
    min_frequency: int,
) -> list[dict[str, object]]:
    """逐月策略：按 bizdate 所在自然月回溯 12 个月窗口，聚合实质元素需求。"""
    if view_pv_count < 0:
        raise ValueError("view_pv_count 不能为负")
    if float(month_total_pv_threshold) < 0:
        raise ValueError("month_total_pv_threshold 不能为负")
    if min_frequency < 0:
        raise ValueError("min_frequency 不能为负")

    bizdate = _validate_partition_dt("bizdate", _partition_yyyymmdd(_today_shanghai()))

    sql = f"""
WITH biz_month AS (
    SELECT
        TO_DATE(CONCAT(SUBSTR('{bizdate}', 1, 4), '-', SUBSTR('{bizdate}', 5, 2), '-01')) AS biz_m1
),
month_window AS (
    SELECT
        CONCAT(
            SUBSTR(CAST(ADD_MONTHS(biz_m1, -12) AS STRING), 1, 4),
            SUBSTR(CAST(ADD_MONTHS(biz_m1, -12) AS STRING), 6, 2)
        ) AS start_ym,
        CONCAT(
            SUBSTR(CAST(ADD_MONTHS(biz_m1, -1) AS STRING), 1, 4),
            SUBSTR(CAST(ADD_MONTHS(biz_m1, -1) AS STRING), 6, 2)
        ) AS end_ym
    FROM biz_month
),
cleaned_video_metrics AS (
    SELECT
        CAST(视频id AS STRING) AS vid,
        SUBSTR(CAST(dt AS STRING), 1, 6) AS ym,
        rov_t0,
        COALESCE(`当日分发曝光pv`, 0) AS day_dist_pv
    FROM loghubods.video_dimension_detail_add_column
    WHERE SUBSTR(CAST(dt AS STRING), 1, 6) >= (SELECT start_ym FROM month_window)
      AND SUBSTR(CAST(dt AS STRING), 1, 6) <= (SELECT end_ym FROM month_window)
      AND COALESCE(`当日分发曝光pv`, 0) >= {int(view_pv_count)}
),
video_monthly_avg_metrics AS (
    SELECT
        ym,
        vid,
        AVG(CASE WHEN rov_t0 = 0 THEN NULL ELSE rov_t0 END) AS vid_avg_rov,
        SUM(day_dist_pv) AS month_total_pv
    FROM cleaned_video_metrics
    GROUP BY ym, vid
    HAVING SUM(day_dist_pv) > {float(month_total_pv_threshold)}
),
tag_vid_dedup AS (
    SELECT DISTINCT
        CAST(vid AS STRING) AS vid,
        原始元素
    FROM loghubods.dwd_topic_decode_result_detail_di
    WHERE dt = MAX_PT('loghubods.dwd_topic_decode_result_detail_di')
      AND 元素维度 = '实质'
      AND 贡献分 >= {float(min_contribution_score)}
),
element_monthly_metrics AS (
    SELECT
        t1.原始元素,
        t2.ym,
        COALESCE(ROUND(AVG(t2.vid_avg_rov), 6), 0) AS month_avg_rov
    FROM tag_vid_dedup t1
    JOIN video_monthly_avg_metrics t2
      ON t1.vid = t2.vid
    GROUP BY t1.原始元素, t2.ym
    HAVING COALESCE(ROUND(AVG(t2.vid_avg_rov), 6), 0) >= {float(rov_avg)}
),
element_total_rov AS (
    SELECT
        原始元素,
        ROUND(SUM(month_avg_rov), 6) AS avg_rov
    FROM element_monthly_metrics
    GROUP BY 原始元素
),
element_vid_dedup AS (
    SELECT DISTINCT
        em.原始元素,
        vm.vid
    FROM element_monthly_metrics em
    JOIN tag_vid_dedup tv
      ON em.原始元素 = tv.原始元素
    JOIN video_monthly_avg_metrics vm
      ON tv.vid = vm.vid
     AND em.ym = vm.ym
),
element_vid_stats AS (
    SELECT
        原始元素,
        COUNT(DISTINCT vid) AS vid_count,
        COLLECT_SET(vid) AS vid_list
    FROM element_vid_dedup
    GROUP BY 原始元素
),
element_freq AS (
    SELECT
        原始元素,
        COUNT(1) AS 频次
    FROM element_monthly_metrics
    GROUP BY 原始元素
)
SELECT
    '逐月' AS strategy,
    md5(CONCAT('逐月', r.原始元素, '{bizdate}')) AS demand_id,
    r.原始元素 AS demand_name,
    r.avg_rov AS weight,
    COALESCE(v.vid_count, 0) AS video_count,
    v.vid_list AS video_list,
    '{{}}' AS ext_info
FROM element_total_rov r
LEFT JOIN element_vid_stats v
  ON r.原始元素 = v.原始元素
LEFT JOIN element_freq f
  ON r.原始元素 = f.原始元素
WHERE r.原始元素 NOT IN (
    '元旦','腊八节','小年','除夕','春节','正月初一','正月初二','正月初三','正月初四','正月初五',
    '情人节','元宵节','龙抬头','妇女节','植树节','劳动节','母亲节','儿童节','端午节','父亲节',
    '建党节','建军节','七夕节','中元节','中秋节','国庆节','重阳节','感恩节','公祭日','平安夜',
    '圣诞节','小寒','大寒','立春','雨水','惊蛰','春分','清明','谷雨','立夏','小满','芒种',
    '夏至','小暑','大暑','立秋','处暑','白露','秋分','寒露','霜降','立冬','小雪','大雪','冬至',
    '早上好','中午好','下午好','晚上好','晚安',
    '祝福','祝愿','祝你','祝贺','祝大家','祝您','祝好运','祝群主','祝朋友'
)
AND COALESCE(f.频次, 0) >= {int(min_frequency)}
ORDER BY r.avg_rov DESC
"""

    odps_client = get_odps_client()
    instance = odps_client.execute_sql(sql)

    rows: list[dict[str, object]] = []
    with instance.open_reader(tunnel=True) as reader:
        for record in reader:
            rows.append(
                {
                    "strategy": record["strategy"],
                    "demand_id": record["demand_id"],
                    "demand_name": record["demand_name"],
                    "weight": _normalize_scalar(record["weight"]),
                    "video_count": record["video_count"],
                    "video_list": _serialize_video_list(record["video_list"]),
                    "ext_info": record["ext_info"],
                }
            )
    return rows