MySql 查询花费的时间比它应该的要长得多

Posted

技术标签:

【中文标题】MySql 查询花费的时间比它应该的要长得多【英文标题】:MySql query taking a lot longer than it should 【发布时间】:2018-04-23 13:03:53 【问题描述】:

由于某种原因,此查询最多需要 5 分钟才能执行。我已将连接缓冲区扩展到 1G,并对此查询进行了说明(结果为 here)似乎没有任何迹象表明这需要这么长时间。在查询期间,所有 8 个 CPU 内核的使用率都达到了接近 100%。

引擎是 InnoDB。

所有表都有一个主键索引。

SELECT Concat(Concat(cust.first_name, ' '), cust.last_name)    AS customerName, 
       TYPE.code                                               AS transType, 
       ty1.nsfamount, 
       np.sumrebateamount, 
       trans.note_id                                           AS note_id, 
       trans.createdate                                        AS createdatestr, 
       n.totalamount, 
       n.currentfloat, 
       ( ( n.costofborrowing * 100 ) / n.amounttolent )        AS fees, 
       n.amounttolent, 
       ( 0 - ( trans.cashamount + trans.chequeamount 
               + trans.debitamount 
               + trans.preauthorizedamount ) )                 AS paidamount, 
       sumpenaltyamount 
FROM   (SELECT * 
        FROM   loan_transaction trans1 
        WHERE  trans1.cashamount < 0 
                OR trans1.chequeamount < 0 
                OR trans1.debitamount < 0 
                OR trans1.preauthorizedamount < 0) trans 
       inner join customer cust 
               ON trans.customer_id = cust.customer_id 
       inner join (SELECT * 
                   FROM   lookuptransactiontypes ty 
                   WHERE  ty.code <> 'REB' 
                          AND ty.code <> 'PN') TYPE 
               ON trans.transactiontype = TYPE.transactiontypesid 
       inner join note n 
               ON trans.note_id = n.note_id 
       inner join (SELECT note_id, 
                          SUM(rebateamount) AS sumrebateamount 
                   FROM   note_payment np1 
                   GROUP  BY np1.note_id) np 
               ON trans.note_id = np.note_id 
       left join (SELECT note_id, 
                         transactiontype, 
                         ( SUM(chequeamount) + SUM(cashamount) 
                           + SUM(debitamount) + SUM(preauthorizedamount) )AS 
                         NSFamount 
                  FROM   (SELECT * 
                          FROM   loan_transaction trans4 
                          WHERE  trans4.cashamount > 0 
                                  OR trans4.chequeamount > 0 
                                  OR trans4.debitamount > 0 
                                  OR trans4.preauthorizedamount > 0)trans5 
                         inner join (SELECT transactiontypesid 
                                     FROM   lookuptransactiontypes ty2 
                                     WHERE  ty2.code = 'NSF')type2 
                                 ON 
                         trans5.transactiontype = type2.transactiontypesid 
                  GROUP  BY trans5.note_id) ty1 
              ON ty1.note_id = trans.refnum 
       left join (SELECT note_id                           AS noteid, 
                         ( SUM(tp.cashamount) + SUM(tp.chequeamount) 
                           + SUM(tp.debitamount) 
                           + SUM(tp.preauthorizedamount) ) AS sumpenaltyamount 
                  FROM   loan_transaction tp 
                         inner join (SELECT transactiontypesid 
                                     FROM   lookuptransactiontypes lp 
                                     WHERE  lp.code = 'PN') lp 
                                 ON lp.transactiontypesid = tp.transactiontype 
                  GROUP  BY tp.note_id) p 
              ON p.noteid = trans.refnum 

最新解释

id  select_type  table       type    possible_keys    key                               key_len  ref                    rows    Extra                            
1   PRIMARY      <derived3>  ALL                                                                                        2241                                     
1   PRIMARY      <derived4>  ALL                                                                                        191441  Using join buffer                
1   PRIMARY      n           eq_ref  PK_NOTE          PK_NOTE                           8        np.note_id             1                                        
1   PRIMARY      <derived2>  ALL                                                                                        274992  Using where; Using join buffer   
1   PRIMARY      cust        eq_ref  PRIMARY_97       PRIMARY_97                        8        trans.CUSTOMER_ID      1                                        
1   PRIMARY      <derived5>  ALL                                                                                        2803                                     
1   PRIMARY      <derived8>  ALL                                                                                        14755                                    
8   DERIVED      <derived9>  ALL                                                                                        2       Using temporary; Using filesort  
8   DERIVED      tp          ref     TRANSACTIONTYPE  TRANSACTIONTYPE                   9        lp.transactionTypesID  110     Using where                      
9   DERIVED      lp          ALL                                                                                        2206    Using where                      
5   DERIVED      <derived7>  ALL                                                                                        98      Using temporary; Using filesort  
5   DERIVED      <derived6>  ALL                                                                                        314705  Using where; Using join buffer   
7   DERIVED      ty2         ALL                                                                                        2206    Using where                      
6   DERIVED      trans4      ALL                                                                                        664587  Using where                      
4   DERIVED      np1         index                    note_payment_idx_id_rebateamount  16                              193366  Using index                      
3   DERIVED      ty          ALL                                                                                        2206    Using where                      
2   DERIVED      trans1      ALL                                                                                        664587  Using where

【问题讨论】:

“似乎没有任何迹象表明为什么这会花费这么多时间”->“使用临时;使用文件排序”表 np1 上的 207662 行是 mysql 中最糟糕的性能问题的指标..relateds此查询SELECT note_id, SUM(rebateamount) AS sumrebateamount FROM note_payment np1 GROUP BY np1.note_id 需要`index(note_id, rebateamount)...换句话说,您需要仔细检查每个交付表所需的索引 刚刚添加了索引,Explain 不再使用文件排序对大量行进行任何操作(解释数据:pastebin.com/q12uHKUE)遗憾的是,查询仍然需要很长时间。 我已经为这个问题添加了最新的解释......是的,如果不知道表结构和索引,很难进一步改进...... 更新:删除 GROUP BY 可将查询速度提高几个数量级 什么版本的 MySQL?较新的版本在JOIN ( SELECT ... ) 上做得更好;你需要这样的优化! 【参考方案1】:

说实话,这个查询有很多问题。您可以按照以下规则轻松简化它:

您可以一次连接多个列(例如:CONCAT(column1, ' ', column2))

无需在同一张表上(或在第一个 FROM 中)执行带有内连接的子查询。只需将您的 FROM 直接放在子查询的主表上,然后将子查询的过滤器移动到主查询的 WHERE 中

对此不确定,但您的所有逻辑似乎都是基于每个 note_id 的。如果确实如此,请在主查询的 GROUP BY => 中移动 GROUP BY note_id 删除每个 note_id 进程执行的所有子查询,只需在想要的表上加入并在主查询中移动它们的 SUM() 和其他列选择查询选择

当您希望有 2 个值基于同一个表但使用不同的过滤器时,您不需要进行子查询,您可以使用 for(比如说 SUM() )示例:

SUM(IF(COLUMN1 = YOUR_FILTER1 OR COLUMN1 = YOUR_FILTER2, COLUMN1, 0)) as totalWithFILTER1andFILTER2 [...] GROUP BY note_id

最后但并非最不重要的一点是,您将加入(内部)在 TYPE.code 上进行过滤的表上,而不是“REB”或“PN”,但随后您将加入(向左)在 TYPE 上进行过滤的结果集上.code = 'PN',这没什么意义,左连接总是会导致NULL

仅供参考,由于我所说的可能看起来很模糊,我开始简化您的查询,但由于我不知道您想要实现什么而停止了这种废话(没有重构 2 LEFT JOIN)。 这是查询(虽然无法测试):

SELECT
  CONCAT(cust.first_name, ' ', cust.last_name) AS customerName,
  TYPE.code                                            AS transType,
  ty1.nsfamount,
  SUM(np.rebateamount) as sumrebateamount,
  trans.note_id                                        AS note_id,
  trans.createdate                                     AS createdatestr,
  n.totalamount,
  n.currentfloat,
  ((n.costofborrowing * 100) / n.amounttolent)         AS fees,
  n.amounttolent,
  (0 - (trans.cashamount + trans.chequeamount
        + trans.debitamount
        + trans.preauthorizedamount))                  AS paidamount,
  sumpenaltyamount
FROM loan_transaction trans
INNER JOIN customer cust ON trans.customer_id = cust.customer_id
INNER JOIN lookuptransactiontypes TYPE ON trans.transactiontype = TYPE.transactiontypesid
INNER JOIN note n ON trans.note_id = n.note_id
INNER JOIN note_payment np ON trans.note_id = np.note_id


LEFT JOIN (SELECT
               note_id,
               transactiontype,
               (SUM(chequeamount) + SUM(cashamount)
                + SUM(debitamount) + SUM(preauthorizedamount)) AS
                 NSFamount
                   FROM loan_transaction trans4
                   INNER JOIN lookuptransactiontypes type2 ON trans4.transactiontype = type2.transactiontypesid
                   WHERE (trans4.cashamount > 0
                         OR trans4.chequeamount > 0
                         OR trans4.debitamount > 0
                         OR trans4.preauthorizedamount > 0) AND type2.code = 'NSF'
             GROUP BY trans5.note_id) ty1
    ON ty1.note_id = trans.refnum
  LEFT JOIN (SELECT
               note_id                         AS noteid,
               (SUM(tp.cashamount) + SUM(tp.chequeamount)
                + SUM(tp.debitamount)
                + SUM(tp.preauthorizedamount)) AS sumpenaltyamount
             FROM loan_transaction tp
               INNER JOIN (SELECT transactiontypesid
                           FROM lookuptransactiontypes lp
                           WHERE lp.code = 'PN') lp
                 ON lp.transactiontypesid = tp.transactiontype
             GROUP BY tp.note_id) p
    ON p.noteid = trans.refnum

WHERE
(trans.cashamount < 0
OR trans.chequeamount < 0
OR trans.debitamount < 0
OR trans.preauthorizedamount < 0)
AND TYPE.code <> 'REB'
AND TYPE.code <> 'PN'

GROUP BY trans.note_id;

【讨论】:

【参考方案2】:

我同意@Aurelien 的回答,为什么您可以加入派生表,而您可以加入普通表并应用过滤器。为什么这样做

-- this will force a full scan on customer table and ignores the filter
select whatever 
from transactions inner join 
(
    select * from customer 
) customer on transactions.customer_id = customer.customer_id
where customer.customer_id  = 1;

虽然你可以这样做

select whatever 
from transactions inner join customer on transactions.customer_id = customer.customer_id
where customer.customer_id  = 1;

除了@Aurelien 的回答,

恕我直言,您查询的问题在于您需要所有客户的数据,因此无论如何优化此查询,您仍然在进行全面扫描,您无法扩展,想象一下几年后你有 1 亿笔交易。

这可能不是您想要的,但是如何分区/分页这样的报告。用户通常不需要一次性展示所有客户,您也不需要浪费资源。

我们的计划是做同样的工作,但只针对 50 位客户。

在您摆脱不必要的子查询后 - 如@Aurelien 回答所示 - 更改他的这部分查询

FROM loan_transaction trans
    INNER JOIN customer cust ON trans.customer_id = cust.customer_id

进入这个

FROM (SELECT * FROM customer LIMIT 50 OFFSET 0) cust 
    INNER JOIN loan_transaction trans ON trans.customer_id = cust.customer_id

请注意,带有偏移的分页不会缩放,因此如果您的客户表很大,您可以考虑another type of pagination

【讨论】:

以上是关于MySql 查询花费的时间比它应该的要长得多的主要内容,如果未能解决你的问题,请参考以下文章

为啥在 C# .NET 中写入 Excel 范围所需的时间比预期的要长得多?

boost::this_thread::sleep_for 的睡眠时间比我预期的要长得多。

SSE 程序在 AMD 上比在 Intel 上花费的时间要长得多

查询花费的时间比预期的要长

使用 wordpress 加载文档的长时间延迟

创建实体框架上下文的实例在负载下减慢