Question

我必须执行一个相对非常大的表（80M记录）Invoice_Payment的更新。它应该更新另一个表Invoice_Payment_updated中的数据，该表是行计数中Invoice_Payment的10％-15％。为了说明，请查看以下演示表：

   Invoice_Payment                         Invoice_Payment_updated
   ---------------                         -----------------------                    
   Customer_id   Invoice_no                Id   Cust_id   Invoice_no      
     10          10100001                  1     10       20200100        
     11          10100002                  2     11       20200101        
     12          10100003
     13          10100004

我知道Merge通常用于执行UPSERT，执行时间比等效的Update语句要长几倍。但相比之下，在某些情况下，具有多个子查询的正常更新语句会降低性能。

MERGE INTO Invoice_Payment ip
USING (SELECT ipu.Cust_id, ipu.Invoice_no from Invoice_Payment_updated ipu
       INNER JOIN Invoice_Payment ip ON ip.Customer_id = ipu.Cust_id
       WHERE ipu.Cust_id = ip.Customer_id and ipu.Invoice_no <> ip.Invoice_no) t
ON (ip.Customer_id = t.Cust_id)
WHEN MATCHED THEN
UPDATE SET ip.Invoice_no = t.Invoice_no;

为了提高性能，我可以使用ROWCOUNT批量更新，但这不会加快执行速度，只会减少整体锁定。

遵循返回相同输出的简单Update语句：

UPDATE Invoice_Payment 
SET Invoice_no = (SELECT ipu.Invoice_no
                  FROM Invoice_Payment_updated ipu
                  WHERE ipu.Cust_id = Invoice_Payment.Customer_id 
                  AND ipu.Invoice_no <> Invoice_Payment.Invoice_no)

WHERE EXISTS (SELECT 1
              FROM Invoice_Payment_updated ipu
              WHERE ipu.Cust_id = Invoice_Payment.Customer_id
              AND ipu.Invoice_no <> Invoice_Payment.Invoice_no);

使用SQL Merge和Update的想法非常聪明，但我听说当我需要在一个大而宽的表中更新许多记录（即超过75M）时，它们都会出现性能问题。此外，重新创建完整的表是很多IO负载，更不用说它会占用大量空间，因为使用子查询，基本上会暂时存储表几次。

使用临时表解决此问题的另一种方法：

CREATE TABLE tmp (
    Cust_id int,
    Invoice_no int);

INSERT INTO tmp_stage VALUES
(SELECT ipu.Cust_id, ipu.Invoice_no FROM Invoice_Payment_updated ipu
 INNER JOIN Invoice_Payment ip ON ip.Customer_id = ipu.Cust_id
 WHERE ipu.Cust_id = ip.Customer_id and ipu.Invoice_no <> ip.Invoice_no);

UPDATE (SELECT tmp.Cust_id, ip.Customer_id, tmp.Invoice_no, tgt.Invoice_no
        FROM tmp INNER JOIN Invoice_Payment ip
        ON tmp.Cust_id = ip.Customer_id)
SET tmp.Invoice_no = ip.Invoice_no;

我想弄清楚在多个子查询的情况下最好使用哪一个？

欢迎任何想法，非常感谢原始问题的完全不同的解决方案。

Answer 1

UPDATE i
SET i.Invoice_no = io.Invoice_no
FROM Invoice_Payment i
   INNER JOIN Invoice_Payment_updated io on i.Customer_id = io.cust_id
WHERE i.Invoice_no <> iu.Invoice_no    -- assuming Invoice_no cannot be NULL

如果更新花费太多时间，请添加WHILE循环和update TOP (10000)直到@@ROWCOUNT = 0。批处理模式可以提高性能。

提高更新大表内容的查询性能的最佳方法

1 个答案: