我正在尝试提高 Oracle SQL 的性能，它正在寻找两个表之间的差异

Question

我有两个 Oracle 表，我在它们之间进行 UNION 以找出存储在这两个表中的数据的差异，但是当我运行 SQL Developer 中的查询然后查询太慢，我在 Informatica 中使用相同的查询，它的吞吐量也更少。

TABLE 1: W_SALES_INVOICE_LINE_FS EBS(NET_AMT, INVOICED_QTY, CREATED_ON_DT, CHANGED_ON_DT, INTEGRATION_ID, 'EBS' 作为 SOURCE_NAME)

TABLE 2: W_SALES_INVOICE_LINE_F DWH (NET_AMT, INVOICED_QTY, CREATED_ON_DT, CHANGED_ON_DT, INTEGRATION_ID, 'EBS' 作为 SOURCE_NAME)

我将问题附加到查询中：

SELECT EBS.NET_AMT, 
nvl(EBS.INVOICED_QTY,
case nvl(EBS.NET_AMT,0) when 0 then EBS.INVOICED_QTY
else -1 end) INVOICED_QTY,
EBS.CREATED_ON_DT,
 EBS.CHANGED_ON_DT, 
     EBS.INTEGRATION_ID,
 'EBS' AS SOURCE_NAME
 FROM
 W_SALES_INVOICE_LINE_FS EBS
  WHERE NOT EXISTS (SELECT  INTEGRATION_ID FROM      W_SALES_INVOICE_LINE_F    DWH
  WHERE EBS.INTEGRATION_ID = DWH.INTEGRATION_ID)
UNION
 SELECT DWH.NET_AMT,
  DWH.INVOICED_QTY, 
  DWH.CREATED_ON_DT,
  DWH.CHANGED_ON_DT, 
  DWH.INTEGRATION_ID,
 'DWH' AS SOURCE_NAME
   FROM
 W_SALES_INVOICE_LINE_F DWH
 where DWH.IS_POS = 'N' and
  not exists (SELECT  INTEGRATION_ID FROM W_SALES_INVOICE_LINE_FS EBS
 WHERE EBS.INTEGRATION_ID = DWH.INTEGRATION_ID);

如果您想查看解释计划，请告诉我。有人可以告诉我如何提高性能，或者让我知道问题是否出在其他方面而不是上述查询！

Answer 1

你不是在表演JOIN，你在表演UNION。但是，您正在执行子查询，而这些可能会降低整体性能。您可以将 EXISTS 更改为 IN，这样可以利用索引（如果存在）。

尝试以下操作：

SELECT EBS.NET_AMT, 
nvl(EBS.INVOICED_QTY,
case nvl(EBS.NET_AMT,0) when 0 then EBS.INVOICED_QTY
else -1 end) INVOICED_QTY,
EBS.CREATED_ON_DT,
 EBS.CHANGED_ON_DT, 
     EBS.INTEGRATION_ID,
 'EBS' AS SOURCE_NAME
 FROM
 W_SALES_INVOICE_LINE_FS EBS
  WHERE EBS.INTEGRATION_ID NOT IN (
     SELECT  INTEGRATION_ID 
     FROM W_SALES_INVOICE_LINE_F
)
UNION ALL
 SELECT DWH.NET_AMT,
  DWH.INVOICED_QTY, 
  DWH.CREATED_ON_DT,
  DWH.CHANGED_ON_DT, 
  DWH.INTEGRATION_ID,
 'DWH' AS SOURCE_NAME
   FROM
 W_SALES_INVOICE_LINE_F DWH
 where DWH.IS_POS = 'N' 
 and DWH.INTEGRATION_ID not in (
     SELECT  INTEGRATION_ID 
     FROM W_SALES_INVOICE_LINE_FS
  );

此外，正如其他人在评论中提到的，UNION ALL 可能更合适。

此外，您可以尝试使用 LEFT OUTER JOIN，如果您有索引，这是执行上述操作的更明确的方法。我无法从我的当前位置访问我的 oracle 来尝试解释计划，但上面和下面实际上可能会进行类似的优化。

SELECT EBS.NET_AMT, 
    Nvl(EBS.INVOICED_QTY,
        CASE Nvl(EBS.NET_AMT, 0) WHEN 0 
        THEN EBS.INVOICED_QTY
        ELSE -1 END
    ) AS INVOICED_QTY,
    EBS.CREATED_ON_DT,
    EBS.CHANGED_ON_DT, 
    EBS.INTEGRATION_ID,
    'EBS' AS SOURCE_NAME
FROM W_SALES_INVOICE_LINE_FS EBS
LEFT OUTER JOIN W_SALES_INVOICE_LINE_F DWH
ON DWH.INTEGRATION_ID = EBS.INTEGRATION_ID
WHERE DWH.INTEGRATION_ID IS NULL
UNION ALL
SELECT DWH.NET_AMT,
    DWH.INVOICED_QTY, 
    DWH.CREATED_ON_DT,
    DWH.CHANGED_ON_DT, 
    DWH.INTEGRATION_ID,
    'DWH' AS SOURCE_NAME
FROM W_SALES_INVOICE_LINE_F DWH
LEFT OUTER JOIN W_SALES_INVOICE_LINE_FS EBS
ON EBS.INTEGRATION_ID = DWH.INTEGRATION_ID
WHERE EBS.INTEGRATION_ID IS NULL
AND DWH.IS_POS = 'N'
;

能否简要说明问题中的 table？每个 table 中有多少（大约）条记录？你有什么指标吗？是否有任何字段 calculated/derived？当您对这些或您的原始查询执行解释计划时，它在哪里显示瓶颈？

Answer 2

不存在和不在语句中通常是性能瓶颈。解决这个问题的一个性能技巧是使用 LEFT OUTER JOIN 和一个声明第二个 table 列为空的子句，即没有匹配的行。所以尝试：

SELECT EBS.NET_AMT, 
nvl(EBS.INVOICED_QTY,
case nvl(EBS.NET_AMT,0) when 0 then EBS.INVOICED_QTY
else -1 end) INVOICED_QTY,
EBS.CREATED_ON_DT,
 EBS.CHANGED_ON_DT, 
     EBS.INTEGRATION_ID,
 'EBS' AS SOURCE_NAME
 FROM
 W_SALES_INVOICE_LINE_FS EBS
 LEFT OUTER JOIN 
  W_SALES_INVOICE_LINE_F    DWH
  ON EBS.INTEGRATION_ID = DWH.INTEGRATION_ID
  WHERE DWH.INTEGRATION_ID IS NULL
UNION
 SELECT DWH.NET_AMT,
  DWH.INVOICED_QTY, 
  DWH.CREATED_ON_DT,
  DWH.CHANGED_ON_DT, 
  DWH.INTEGRATION_ID,
 'DWH' AS SOURCE_NAME
   FROM
 W_SALES_INVOICE_LINE_F DWH
 LEFT OUTER JOIN W_SALES_INVOICE_LINE_FS EBS
 ON EBS.INTEGRATION_ID = DWH.INTEGRATION_ID
 where EBS.INTEGRATION_ID IS NULL
 AND DWH.IS_POS = 'N'

Answer 3

您正在手动编写完整的外部联接，Oracle 可以自动为此类比较任务执行此操作（我猜它可能运行更快）

select
ebs.net_amt ebs_net_amt,
dwh.net_amt dwh_net_amt,
nvl(ebs.invoiced_qty,case nvl(ebs.net_amt,0) when 0 then ebs.invoiced_qty else -1 end) invoiced_qty_ebs,
dwh.invoiced_qty invoiced_qty_dwh,
ebs.created_on_dt ebs_created_on_dt,
dwh.created_on_dt dwh_created_on_dt,
ebs.changed_on_dt ebs_changed_on_dt,
dwh.changed_on_dt dwh_changed_on_dt,
nvl(ebs.integration_id,ebs.integration_id) integration_id,
case
when ebs.integration_id is not null and ebs.integration_id is not null and then 'EBS and DWH'
when dwh.integration_id is not null then 'EBS'
else 'DWH' 
end source_name
from
w_sales_invoice_line_fs ebs
full outer join
(select * from w_sales_invoice_line_f dwh where dwh.is_pos = 'N') dwh
on 
(ebs.integration_id = dwh.integration_id)
where
ebs.integration_id is null or dwh.integration_id is null --restrict to records missing on one side

我正在尝试提高 Oracle SQL 的性能，它正在寻找两个表之间的差异

I am trying to improve the performance of an Oracle SQL that is finding the differences between two tables

sql

oracle

informatica