跨不同事件/行的密钥/值对串联和汇总键/值对

Question

我有一个由客户事件组成的表。这些事件具有包含密钥和值的数组。

以下模拟表说明了这些事件的外观：

事件表

WITH
  CTE_TEST_DATA AS (
    SELECT
      'XXX'     AS user
      ,'2025-01-01' AS event_date
      ,'store_1'    AS store
      ,'london'     AS city
      ,[
          STRUCT('type' AS key, NULL AS int_value, NULL as float_value, 'online' AS string_value),
          STRUCT('ware', NULL, NULL, 'banana'),
          STRUCT('amount', 2, NULL, NULL),
          STRUCT('price', NULL, 0.5, NULL)
        ] AS purchases
    UNION ALL
    SELECT
      'XXX'     AS user
      ,'2025-01-01' AS event_date
      ,'store_1'    AS store
      ,'london'     AS city
      ,[
          STRUCT('type', NULL, NULL, 'online'),
          STRUCT('ware', NULL, NULL, 'apple'),
          STRUCT('amount', 3, NULL, NULL),
          STRUCT('price', NULL, 0.6, NULL)
        ] AS purchases
    UNION ALL
    SELECT
      'XXX'     AS user
      ,'2025-01-01' AS event_date
      ,'store_1'    AS store
      ,'london'     AS city
      ,[
          STRUCT('type', NULL, NULL, 'online'),
          STRUCT('ware', NULL, NULL, 'banana'),
          STRUCT('amount', 3, NULL, NULL),
          STRUCT('price', NULL, 0.5, NULL)
        ] AS purchases
    UNION ALL
    SELECT
      'XXX'     AS user
      ,'2025-01-02' AS event_date
      ,'store_1'    AS store
      ,'london'     AS city
      ,[
          STRUCT('type', NULL, NULL, 'online'),
          STRUCT('ware', NULL, NULL, 'bread'),
          STRUCT('amount', 1, NULL, NULL),
          STRUCT('price', NULL, 1.0, NULL)
        ] AS purchases
    UNION ALL
    SELECT
      'YYY'     AS user
      ,'2025-01-01' AS event_date
      ,'store_2'    AS store
      ,'sydney'     AS city
      ,[
          STRUCT('type', NULL, NULL, 'physical'),
          STRUCT('ware', NULL, NULL, 'milk'),
          STRUCT('amount', 1, NULL, NULL),
          STRUCT('price', NULL, 1.5, NULL)
        ] AS purchases
    )

实际客户事件表非常大，因此已经创建了每天的聚合查询，以使分析和报告成本降低。

本每日聚合查询每次用户每次用户的汇总结果_date：

SELECT
   CTD.user
  ,CTD.event_date
  ,STRUCT(COUNT(*) as amount) AS events_per_user
  ,*𝑜𝑡ℎ𝑒𝑟 𝑎𝑔𝑔𝑟𝑒𝑔𝑎𝑡𝑒𝑠*
FROM
  CTE_TEST_DATA  AS CTD
GROUP BY
   CTD.user
  ,CTD.event_date

当新指标添加到客户事件表中时，必须对每日聚合查询进行修改以包括这些查询。

由于此设置的问题，我在添加汇总列时有一些约束，主要的是必须使用相关的子查询专门添加新指标，并且不能使用常见的表格表达式。

我试图实现的结果是以下内容，其中数组仅包含不同的商品名称：

用户 event_dateevents_per_user.amountware.nameware.distinct_orders2025-01-013111xxx2025-01-021bread11在此表中导致该表：

		total.amount			xxx
bana	2	5
apple	1	3		Yyy	2025-01-01
米尔克
				不幸的是，我最接近实现该结果的是使用以下查询：	SELECT CTD.user ,CTD.event_date ,STRUCT(COUNT(*) as amount) AS events_per_user ,ARRAY_AGG(( SELECT AS STRUCT P.string_value AS name ,COUNT(P.string_value) AS distinct_orders FROM UNNEST(CTD.purchases) AS P WHERE 1=1 AND P.key = 'ware' GROUP BY P.string_value )) AS ware ,ARRAY_AGG(( SELECT AS STRUCT SUM(P.int_value) AS amount FROM UNNEST(CTD.purchases) AS P WHERE 1=1 AND P.key = 'amount' GROUP BY user ,event_date )) AS total FROM CTE_TEST_DATA AS CTD GROUP BY CTD.user ,CTD.event_date

用户

event_date

events_per_user.amount

ware.nameware.distinct_ordersbana11111xxx2025-01-021bread11任何帮助将不胜感激！

WITH
  CTE_TEST_DATA AS (...),
  exploded_data AS (
    -- Decompose the structure into separate lines
    SELECT
      user,
      event_date,
      store,
      city,
      (
        SELECT AS STRUCT * 
        FROM UNNEST(purchases) AS p 
        ORDER BY p.key DESC LIMIT 1
      ) AS ware,
      (
        SELECT AS STRUCT * 
        FROM UNNEST(purchases) AS p 
        ORDER BY p.key DESC LIMIT 1 OFFSET 1
      ) AS type,
      (
        SELECT AS STRUCT * 
        FROM UNNEST(purchases) AS p 
        ORDER BY p.key DESC LIMIT 1 OFFSET 2
      ) AS price,
      (
        SELECT AS STRUCT * 
        FROM UNNEST(purchases) AS p 
        ORDER BY p.key DESC LIMIT 1 OFFSET 3
      ) AS amount
    FROM
      CTE_TEST_DATA
  ),
  aggregated_fruit_lvl AS (
    -- Aggregate data on user/day/fruit level
    SELECT
      user,
      event_date,
      ware.string_value,
      COUNT(ware.string_value) AS distinct_count,
      SUM(amount.int_value) AS sum_count
    FROM
      exploded_data
    GROUP BY
      user,
      event_date,
      ware.string_value
  ),
  aggregated_day_lvl AS (
    -- Calculate the number of transactions per day by user
    SELECT
      user,
      event_date,
      COUNT(1) AS total
    FROM
      exploded_data
    GROUP BY
      user,
      event_date
  ),
  aggregated_fruit_lvl_with_struct AS (
    -- Create the structure: fruit / count / sum purchase
    SELECT
      user,
      event_date,
      ARRAY_AGG(
        STRUCT(string_value, distinct_count, sum_count)
      ) AS name_number_list
    FROM
      aggregated_fruit_lvl
    GROUP BY
      user,
      event_date
  )
SELECT
  aflws.user,
  aflws.event_date,
  adl.total,
  aflws.name_number_list
FROM
  aggregated_fruit_lvl_with_struct aflws
JOIN
  aggregated_day_lvl adl ON aflws.user = adl.user
  AND aflws.event_date = adl.event_date;

total.amount			xxx	2025-01-01	3
2					apple
3					bana
3		Yyy	2025-01-01	1	米尔克
		不幸的是，我花了数小时的时间研究和尝试不同的方法来实现这一结果，但我不知道是否有可能。			为了解决该问题，我首先使用Un -nest将数据归一化以改变现有结构。在此步骤之后，我通过重建所需的结构来执行聚合并重建最终数据：