Jooq / SQL查找另一列中唯一值的平均值

Question

我有一个查询从表的组合返回各种数字。我正在使用jooq来运行此查询。

final SiteSalesFigures siteSalesFigures =
dsl.select(
      countDistinct(LINE.TRANSACTION_ID).as("transactionCount"),
      sum(LINE.PROFIT).as("totalProfit"),
      sum(LINE.TOTAL).as("totalSalesAmount"),
      sum(LINE.QUANTITY).as("totalItemsSold"),
      sum(LINE.PROFIT).divide(sum(LINE.TOTAL)).multiply(100).as("profitMarginPercentage"),
      avg(TRANSACTIONS.NO_OF_ITEMS).as("averageItemsPerTransaction"),
      sum(LINE.TOTAL).divide(countDistinct(LINE.TRANSACTION_ID)).as("averageSalesTotalPerTransaction"),
      sum(LINE.PROFIT).divide(countDistinct(LINE.TRANSACTION_ID)).as("averageProfitTotalPerTransaction"))
    .from(TRANSACTIONS)
    .join(LINE).on(TRANSACTIONS.TRANSACTION_ID.equal(LINE.TRANSACTION_ID))
    .leftJoin(ITEM).on(LINE.ITEM_ID.equal(ITEM.ITEM_CODE))
    .where(TRANSACTIONS.SITE_ID.equal(siteId))
    .and(TRANSACTIONS.NO_OF_LINES.greaterThan(0))
    .and(TRANSACTIONS.START_TIME
      .between(new Timestamp(reportStartDate.toInstant().toEpochMilli()))
      .and(new Timestamp(reportEndDate.toInstant().toEpochMilli())))
    .and(TRANSACTIONS.TRANSACTION_TYPE_ID.notEqual(cancelledSaleID))
    .fetchOneInto(SiteSalesFigures.class);

averageItemsPerTransaction被证明是个问题。我完全理解为什么它不起作用，但我不确定如何才能使它工作。不幸的是，由于使用Line表的排除，因此必须进行连接。

如果事务有3行，则事务详细信息（包括no_of_items）将被复制三次，这会导致值不正确。

我知道正确的值，因为我只在事务表上运行了平均（no_of_items）查询。

以下是仅两个事务的表格（隐藏此示例不需要的列）：

          **transaction_id**       **no_of_lines    no_of_items**
8abf1720-51f6-a1bf-4714-004b644cb99f --- 2         --- 2
8abf1720-51f6-a1bf-4714-004b644cb99f --- 2         --- 2
d239feab-38ea-7c8a-4814-7d5a38f74949 --- 3         --- 4
d239feab-38ea-7c8a-4814-7d5a38f74949 --- 3         --- 4
d239feab-38ea-7c8a-4814-7d5a38f74949 --- 3         --- 4

您会注意到行数并不总是等于项目数（例如，一行可以扫描两个项目）

有人有解决方案吗？

Answer 1

Solution using 2 queries

一个明显的解决方案是运行两个查询来获得这些结果。第一个查询将是您已经拥有的（但没有平均值），第二个查询只计算平均值：

final SiteSalesFigures siteSalesFigures =
dsl.select(
      avg(TRANSACTIONS.NO_OF_ITEMS).as("averageSalesTotalPerTransaction"),
      avg(TRANSACTIONS.PRICE).as("averageSalesTotalPerTransaction"),
      avg(TRANSACTIONS.PROFIT).as("averageProfitTotalPerTransaction"))
    .from(TRANSACTIONS)
    .where(TRANSACTIONS.SITE_ID.equal(siteId))
    .and(TRANSACTIONS.NO_OF_LINES.greaterThan(0))
    .and(TRANSACTIONS.START_TIME
      .between(new Timestamp(reportStartDate.toInstant().toEpochMilli()))
      .and(new Timestamp(reportEndDate.toInstant().toEpochMilli())))
    .and(TRANSACTIONS.TRANSACTION_TYPE_ID.notEqual(cancelledSaleID))
    .fetchOneInto(SiteSalesFigures.class);

这可能比一次性完成所有速度要慢，具体取决于TRANSACTIONS表的大小。

Solution using a weighted average

因为您的连接会生成重复的TRANSACTIONS行，所以您必须计算加权平均值，而不是普通平均值。举个例子，如果你的TRANSACTIONS行被重复3次，那么你必须将该特定事务的贡献除以3.这通常会非常复杂，但考虑到你已经通过预先计算每个事务的NO_OF_ITEMS来对你的模式进行非规范化你很幸运如果您没有此列，则必须在派生表中预先计算它。

在SQL / jOOQ中：

final SiteSalesFigures siteSalesFigures =
dsl.select(
      ...
      count() 
        .divide(countDistinct(TRANSACTIONS.TRANSACTION_ID)).as("averageSalesTotalPerTransaction"),
      sum(TRANSACTIONS.PRICE.divide(TRANSACTIONS.NO_OF_ITEMS))
        .divide(countDistinct(TRANSACTIONS.TRANSACTION_ID)).as("averageSalesTotalPerTransaction"),
      sum(TRANSACTIONS.PROFIT.divide(TRANSACTIONS.NO_OF_ITEMS))
        .divide(countDistinct(TRANSACTIONS.TRANSACTION_ID)).as("averageProfitTotalPerTransaction"))
    .from(TRANSACTIONS)
    .join(...)
    ...
    .fetchOneInto(SiteSalesFigures.class);

根据您的数据类型，您可能需要转换为DOUBLE或NUMBER。

I've blogged about calculating weighted averages in SQL more in detail here。

Answer 2

解决方案一直在我面前，我可以使用正在工作的值来获取我需要的值：

final SiteSalesFigures siteSalesFigures =
dsl.select(
      countDistinct(LINE.TRANSACTION_ID).as("transactionCount"),
      sum(LINE.PROFIT).as("totalProfit"),
      sum(LINE.TOTAL).as("totalSalesAmount"),
      sum(LINE.QUANTITY).as("totalItemsSold"),
      sum(LINE.PROFIT).divide(sum(LINE.TOTAL)).multiply(100).as("profitMarginPercentage"),
      sum(LINE.QUANTITY).divide(countDistinct(LINE.TRANSACTION_ID)).as("averageItemsPerTransaction"),
      sum(LINE.TOTAL).divide(countDistinct(LINE.TRANSACTION_ID)).as("averageSalesTotalPerTransaction"),
      sum(LINE.PROFIT).divide(countDistinct(LINE.TRANSACTION_ID)).as("averageProfitTotalPerTransaction"))
    .from(TRANSACTIONS)
    .join(LINE).on(TRANSACTIONS.TRANSACTION_ID.equal(LINE.TRANSACTION_ID))
    .leftJoin(ITEM).on(LINE.ITEM_ID.equal(ITEM.ITEM_CODE))
    .where(TRANSACTIONS.SITE_ID.equal(siteId))
    .and(TRANSACTIONS.NO_OF_LINES.greaterThan(0))
    .and(TRANSACTIONS.START_TIME
      .between(new Timestamp(reportStartDate.toInstant().toEpochMilli()))
      .and(new Timestamp(reportEndDate.toInstant().toEpochMilli())))
    .and(TRANSACTIONS.TRANSACTION_TYPE_ID.notEqual(cancelledSaleID))
    .fetchOneInto(SiteSalesFigures.class);