Redshift 相关子查询内部错误

Posted 2023-03-31

【中文标题】Redshift 相关子查询内部错误

【英文标题】：Redshift Correlated Subquery error internal

【发布时间】：2020-03-07 07:50:26

【问题描述】：

我在 mysql 中有一个查询，它计算来自特定供应商的 products 和 product_status，如 Live, Pause, soldout, Partial-Soldout 等。查询包括子查询，但在 Mysql 中完美运行。对于 Redshift (Postgre v8.x)，它会给出错误 correlated subquery pattern is not supported due to internal error

查询 (POSTGRES)

SELECT COUNT(CASE WHEN (vendor_id = 6 AND status = 1) THEN 1 ELSE NULL END) AS "vex",
   COUNT(CASE WHEN (vendor_id = 6 AND status = 1 AND p.p_id IN (SELECT pov.p_id FROM product_option_value pov WHERE pov.p_id AND p.quantity != pov.quantity AND pov.quantity = 0 GROUP BY pov.p_id)) THEN 1 ELSE NULL END) AS "vex-Partial-Soldout",
   COUNT(CASE WHEN (vendor_id = 6 AND status = 1 AND p.quantity = 0) THEN 1 ELSE NULL END) AS "vex-Soldout",
   COUNT(CASE WHEN (vendor_id = 5 AND status = 1) THEN 1 ELSE NULL END) AS "vey-DXB",
   COUNT(CASE WHEN (vendor_id = 5 AND status = 1 AND p.p_id IN (SELECT pov.p_id FROM product_option_value pov WHERE pov._id AND p.quantity != pov.quantity AND pov.quantity = 0 GROUP BY pov.p_id)) THEN 1 ELSE NULL END) AS "vey-Partial-Soldout",
   COUNT(CASE WHEN (vendor_id = 5 AND status = 1 AND p.quantity = 0) THEN 1 ELSE NULL END) AS "vey-Soldout"
FROM product p

表结构

//Product p table
*  p_id *   model   *   vendor_id   *   status  * Quantity * 
* 1001  *   HB1     *      1        *   1       *  10      *
* 1002  *   HB2     *      6        *   1       *  17      *
* 1003  *   HB3     *      5        *   1       *  19      *
* 1004  *   HB4     *      2        *   1       *  3       *
* 1005  *   HB5     *      1        *   1       *  8       *
* 1006  *   HB6     *      6        *   1       *  55      *
* 1007  *   HB7     *      3        *   1       *  32      *
* 1008  *   HB8     *      5        *   1       *  6       *
* 1009  *   HB9     *      5        *   1       *  10      *


//product_option_value pov table
*   pov_id  *   p_id    *   opt_id    *   quantity   * 
*   1       *   1001    *   11        *   10         *
*   2       *   1002    *   11        *   17         *
*   3       *   1003    *   11        *   0          *
*   4       *   1004    *   11        *   3          *
*   5       *   1005    *   11        *   8          *
*   6       *   1006    *   11        *   0          *
*   7       *   1007    *   11        *   32         *
*   8       *   1008    *   11        *   6          *
*   9       *   1009    *   11        *   0          *

Group by 在子查询中是必需的，所以左连接也不能解决问题。

【问题讨论】：

***.com/help/minimal-reproducible-example

将查询缩小到 SELECT 子句中的一列，然后缩小表达式，直到找到导致问题的特定元素

仅供参考，将SELECT 语句放在 Select 行中是编写 SQL 查询的一种非常低效的方式。它实际上需要为外部查询的每一行运行一个查询。您可能可以将其重写为使用LEFT OUTER JOIN，然后测试连接的列是否为NULL。

@JohnRotenstein，我是你的忠实粉丝..我关注了你的很多答案......我尝试了LEFT JOIN，但查询需要按产品 ID 分组

您可以对查询（包括 GROUP BY）执行 LEFT OUTER JOIN。

【参考方案1】：

子查询中的逻辑很难理解，所以我不确定我是否 100% 正确。例如，样本数据似乎每个产品只有一行，但我不知道是否真的如此。或者opt_id 是什么，因为这在使用pov 表的查询中似乎很有用。

也就是说，您似乎只需要JOIN 和GROUP BY 就可以得到您想要的——一个更简单且在任何数据库中都应该更快的查询。

在以下查询中，我还将两个供应商拆分到不同的行。这至少有助于使逻辑正确：

SELECT vendor_id,
       COUNT(DISTINCT p.p_id) AS num_products,
       SUM(CASE WHEN p.quantity <> pov.quantity AND pov.quantity = 0 THEN 1 ELSE 0 END) as partial_soldout,
       COUNT(DISTINCT CASE WHEN p.quantity = 0 THEN p.p_id END) as soldout
FROM product p LEFT JOIN
     product_option_value pov
     ON pov.p_id = p.p_pid
WHERE p.vendor_id IN (5, 6) AND p.status = 1
GROUP BY p.vendor_id;

【参考方案2】：

您必须从计算“vex-Partial-Soldout”和“vey-Partial-Soldout”的两个表达式中删除 where 子句之后的“pov.p_id AND”。然后它在 postgreSQL9.6 上工作，输出与您的 sqlserver 版本输出匹配。

SELECT COUNT(CASE WHEN (vendor_id = 6 AND status = 1) THEN 1 ELSE NULL END) AS "vex",
   COUNT(CASE WHEN (vendor_id = 6 AND status = 1 AND p.p_id IN (SELECT pov.p_id FROM product_option_value pov WHERE p.quantity != pov.quantity AND pov.quantity = 0 GROUP BY pov.p_id)) THEN 1 ELSE NULL END) AS "vex-Partial-Soldout",
   COUNT(CASE WHEN (vendor_id = 6 AND status = 1 AND p.quantity = 0) THEN 1 ELSE NULL END) AS "vex-Soldout",
   COUNT(CASE WHEN (vendor_id = 5 AND status = 1) THEN 1 ELSE NULL END) AS "vey-DXB",
   COUNT(CASE WHEN (vendor_id = 5 AND status = 1 AND p.p_id IN (SELECT pov.p_id FROM product_option_value pov WHERE p.quantity != pov.quantity AND pov.quantity = 0 GROUP BY pov.p_id)) THEN 1 ELSE NULL END) AS "vey-Partial-Soldout",
   COUNT(CASE WHEN (vendor_id = 5 AND status = 1 AND p.quantity = 0) THEN 1 ELSE NULL END) AS "vey-Soldout"
FROM product p

注意：我没有过多关注您的逻辑，作为最佳实践，请避免在您的选择中编写选择语句。