SQL查询速度优化

1、使用left join比直接使用where速度快

参考：使用left join比直接使用where速度快的原因

多表使用left join只是把主表里的所有数据查询出来，其他表只查询表中的符合条件的某一条记录，所以速度非常快；而多表使用where内联，是把所有表的数据全查出来，然后进行比对，所以速度非常慢。

使用left join要注意确定哪一张表是主表，如果无法确定主表，则选择哪张表查询的字段最多，就把哪张表作为主表。

示例如下：

使用left join,同样的数据量，时间不到1秒钟！

SELECT a.projectno,MAX(a.projectname) projectname,max(a.projectMoney) projectMoney,
max(a.projectLimitYear) projectLimitYear,max(a.monthcharge) monthcharge,
max(c.orgname) orgname,max(d.businesstypename) businesstypename,max(e.name) name,
max(b.dicvalue) dicvalue,
min(CONVERT(varchar(100),DATEADD("DAY",jbl.DelayDays,jbl.ReportTime),23)) as period,
max(f.fiveleveltype) fiveleveltype,max(a.ProjectInfoId) ProjectInfoId,
max(g.FlowRunId) FlowRunId
FROM
(select ProjectInfoId,ProjectNo,Status,OrgId,TypeID,UserAId,IsDelete,ProjectName,ProjectMoney,ProjectLimitYear,MonthCharge from jt_biz_projectinfo) a
LEFT JOIN
(select b.* from (select MAX(id) id, ProjectNo from JT_Biz_Lecture group by ProjectNo) a,
JT_Biz_Lecture b where a.id=b.id) f
on a.projectno=f.projectno
LEFT JOIN
(select DicId,IsDelete,DicValue,DicCode from jt_Base_Dictionary) b
ON a.status=b.dicId
LEFT JOIN
(select ProjectNo,DelayDays,ReportTime from Jt_Biz_LectureTemp) jbl
ON a.projectno=jbl.projectno
LEFT JOIN
(select IsDelete,OrgName,OrgId from jt_base_org) c
ON a.orgid=c.orgid
LEFT JOIN
(select ID,BusinessTypeName from d_businesstype) d
ON a.typeid=d.id
LEFT JOIN
(select UserInfoId,IsDelete,Name,UserName from jt_base_userInfo) e
ON a.useraid=e.userinfoid
LEFT JOIN
(select UserId,PostId,OrgIds from JT_Base_Post_UserOrg) j
ON j.UserId=e.userinfoid
LEFT JOIN
(select PostId from JT_Base_Post) k
ON j.PostId=k.PostId
LEFT JOIN
(select FlowRunId,ProjectId from jt_flow_run) g
ON a.ProjectInfoId=g.ProjectId
LEFT JOIN
(select FlowRunId from JT_Flow_Run_Prcs) h
ON g.FlowRunId = h.FlowRunId
LEFT JOIN
(select OrgId,OrgName from JT_Base_Org) l
ON a.OrgId=l.OrgId
WHERE a.status in(82,83,84) and a.isdelete=0
and b.isdelete=0 and c.isdelete=0 and e.isdelete=0
and (l.OrgId in(null) or e.UserName='chenqf') group by a.ProjectNo;

不使用left join,同样数据量，时间大概在50秒的样子！
select a.projectno,max(a.projectname) projectname,max(a.projectMoney) projectMoney,
max(a.projectLimitYear) projectLimitYear,max(a.monthcharge) monthcharge,
max(c.orgname) orgname,max(d.businesstypename) businesstypename,max(e.name) name,
max(b.dicvalue) dicvalue,
min(CONVERT(varchar(100),DATEADD("DAY",jbl.DelayDays,jbl.ReportTime),23)) as period,
max(f.fiveleveltype) fiveleveltype,max(a.ProjectInfoId) ProjectInfoId,
max(g.FlowRunId) FlowRunId
from (select DicId,IsDelete,DicValue,DicCode from jt_Base_Dictionary) b,
(select IsDelete,OrgName,OrgId from jt_base_org) c,
(select UserInfoId,IsDelete,Name,UserName from jt_base_userInfo) e,
(select ID,BusinessTypeName from d_businesstype) d,
(select FlowRunId,ProjectId from jt_flow_run) g,
(select FlowRunId from JT_Flow_Run_Prcs) h,
(select UserId,PostId,OrgIds from JT_Base_Post_UserOrg) j,
(select PostId from JT_Base_Post) k,
(select OrgId,OrgName from JT_Base_Org) l,
(select ProjectNo,DelayDays,ReportTime from Jt_Biz_LectureTemp) jbl,
(select ProjectInfoId,ProjectNo,Status,OrgId,TypeID,UserAId,IsDelete,ProjectName,ProjectMoney,ProjectLimitYear,MonthCharge from jt_biz_projectinfo) a
left join
(select b.* from (select MAX(id) id, ProjectNo from JT_Biz_Lecture group by ProjectNo) a,
JT_Biz_Lecture b where a.id=b.id) f
on a.projectno=f.projectno where a.status=b.dicId and a.status in(82,83,84)
and a.projectno=jbl.projectno and a.orgid=c.orgid and a.typeid=d.id
and a.useraid=e.userinfoid and j.UserId=e.userinfoid and a.isdelete=0
and b.isdelete=0 and c.isdelete=0 and e.isdelete=0 and g.FlowRunId = h.FlowRunId 
and (l.OrgId in(null) or e.UserName='chenqf')
group by a.ProjectNo;

2、LEFT JOIN关联表中ON,WHERE后面跟条件的区别

写SQL本想通过 A left join B on and 后面的条件来使查出的两条记录变成一条，奈何发现还是有两条。

后来发现  on and 不会过滤结果记录条数，只会根据and后的条件是否显示 B表的记录，A表的记录一定会显示。

不管and 后面的是A.id=1还是B.id=1,都显示出A表中所有的记录，并关联显示B中对应A表中id为1的记录或者B表中id为1的记录。

运行sql : select * from student s left join class c on s.classId=c.id order by s.id

运行sql : select * from student s left join class c on s.classId=c.id and s.name="张三" order by s.id

运行sql : select * from student s left join class c on s.classId=c.id and c.name="三年级三班" order by s.id

       通过查询相关资料了解到：

       数据库在通过连接两张或多张表来返回记录时，都会生成一张中间的临时表，然后再将这张临时表返回给用户。

      在使用left jion时，on和where条件的区别如下：

1、 on条件是在生成临时表时使用的条件，它不管on中的条件是否为真，都会返回左边表中的记录。

2、where条件是在临时表生成好后，再对临时表进行过滤的条件。这时已经没有left join的含义（必须返回左边表的记录）了，条件不为真的就全部过滤掉。

       假设有两张表：

表1：tab1

表2：tab2

两条SQL:
1、select * form tab1 left join tab2 on (tab1.size = tab2.size) where tab2.name=’AAA’
2、select * form tab1 left join tab2 on (tab1.size = tab2.size and tab2.name=’AAA’)

    

     其实以上结果的关键原因就是left join,right join,full join的特殊性，不管on上的条件是否为真都会返回left或right表中的记录，full则具有left和right的特性的并集。 而inner jion没这个特殊性，则条件放在on中和where中，返回的结果集是相同的。

3、Where条件中使用主表条件过滤速度较快

		SELECT
       *
		FROM
			USER_BORROW T1   -- 数据量2~3万条
		LEFT JOIN USER T3 ON T3.OID_USER_ID = T1.OID_USER_ID
		LEFT JOIN USER_DETAIL T4 ON T4.OID_USER_ID = T1.OID_USER_ID
		LEFT JOIN USER_ARCHIVES T5 ON T5.OID_ARCHIVE_ID = T1.OID_ARCHIVE_ID    -- USER_ARCHIVES数据量2~3万条
		LEFT JOIN PRODUCT_MST T6 ON T6.OID_PROD_ID = T5.OID_PROD_ID
		LEFT JOIN TENANT T7 ON T7.OID_TENANT_ID = T5.OID_TENANT_ID
		LEFT JOIN TENANT T8 ON T8.OID_TENANT_ID = T5.OID_TENANT_XS_ID
		LEFT JOIN (
			SELECT
				OID_BORROW_ID,
				SUM(REPAY_AMOUNT_TOTAL) AS FEE_AMOUNT
			FROM
				USER_POUNDAGE_REPAY
			WHERE
				REPAY_FLG = '0'
			GROUP BY
				OID_BORROW_ID
		) T9 ON T1.BORROW_ID = T9.OID_BORROW_ID
		WHERE
			1 = 1
     -- 此处使用T1.OID_TENANT_ID进行数据过滤查询速度很快，
     -- 但是使用T5.OID_TENANT_ID进行过滤数据查询速度极慢，不确定是否跟T1为主表有关系？？？
		AND T1.OID_TENANT_ID IN (  
			SELECT
				OT.OID_TENANT_ID
			FROM
				DEPARTMENT_USER DU
			INNER JOIN OPERATOR_TENANT OT ON OT.OID_OPERATOR_USER_ID = DU.OID_ADMIN_USER_ID
			WHERE
				DU.DEL_FLG = '0'
			AND DU.OID_DEPARTMENT_ID = (
				SELECT
					DU1.OID_DEPARTMENT_ID
				FROM
					DEPARTMENT_USER DU1
				WHERE
					DU1.OID_ADMIN_USER_ID = ?
				AND DU1.DEL_FLG = '0'
			)
		)