为什么在 Oracle 中使用 CASE 进行子查询比 JOIN WITH OR 更快

Question

我正在优化我们拥有的一个可怕的视图，令人惊讶的是，使用 CASE 语句的子选择之一 运行 比使用 OR 的 LEFT JOIN 快。原始视图大得多，但我感兴趣的部分可以归结为以下查询

SELECT CASE 
     WHEN tdcurr.productid = 1 THEN (SELECT addressid 
                                     FROM   address a 
                                     WHERE  a.customerid = tm.customerid 
                                            AND a.addressid = 
                                                tdcurr.addressid 
                                            AND a.addresstypeid = 3) 
     WHEN tdcurr.productid = 2 THEN (SELECT addressid 
                                     FROM   address a 
                                     WHERE  a.customerid = tm.customerid 
                                            AND a.addressid = 
                                                tdcurr.addressid 
                                            AND a.addresstypeid = 4) 
   END AS t_buyselladdressid 
FROM   vleaf_transactiondetail_all tdcurr 
   inner join transactionmain tm 
           ON tm.transactionid = tdcurr.transactionid 

执行计划

而使用 join 的速度一直较慢

SELECT bsaddr.addressid AS t_buyselladdressid 
FROM   vleaf_transactiondetail_all tdcurr 
   inner join transactionmain tm 
           ON tm.transactionid = tdcurr.transactionid 
   left outer join address bsaddr 
                ON tm.customerid = bsaddr.customerid 
                   AND bsaddr.addressid = tdcurr.addressid 
                   AND ( ( tdcurr.productid = 1 
                           AND bsaddr.addresstypeid = 3 ) 
                          OR ( tdcurr.productid = 2 
                               AND bsaddr.addresstypeid = 4 ) ) 

执行计划

为什么会这样？

Answer 1

带有子选择的 SQL 可能受益于标量子查询缓存。从解释计划来看，它看起来确实受益于不执行嵌套循环外连接！

有关标量子查询缓存的详细信息，请参阅 https://asktom.oracle.com/pls/asktom/f?p=100:11:0::::P11_QUESTION_ID:2683853500346598211。