22,300
社区成员
发帖
与我相关
我的任务
分享SQLSERVER Union效率难题求解
遇到一个奇怪现象，百思不得奇解
sql是这样的：
第一个查询
with (树 a)
select count(*) from b where b.code = a.code and (b的过滤条件）
耗时0.x秒
第二个查询
with (树 a)
select count(*) from c where c.code = a.code and (c的过滤条件）
耗时也是0.x秒
但这样写就出问题了
with (树 a)
select count(*) from b where b.code = a.code and (b的过滤条件）
union
select count(*) from c where c.code = a.code and (c的过滤条件）
耗时 60+秒......
表数据 a 和 b大的一个是在50万行而已
检查执行路径，似乎c表的全表检索耗时71%，但看不出什么问题来
sql是这样的：
第一个查询
with (树 a)
select count(*) from b where b.code = a.code and (b的过滤条件）
耗时0.x秒
第二个查询
with (树 a)
select count(*) from c where c.code = a.code and (c的过滤条件）
耗时也是0.x秒
但这样写就出问题了
with (树 a)
select count(*) from b where b.code = a.code and (b的过滤条件）
union
select count(*) from c where c.code = a.code and (c的过滤条件）
耗时 60+秒......
表数据 a 和 b大的一个是在50万行而已
检查执行路径，似乎c表的全表检索耗时71%，但看不出什么问题来
...全文
请发表友善的回复…
发表回复
BurningM 2010-02-25
- 打赏
- 举报
试了下，在referencecount上加了个索引，果然查询效率高了很多
但不在union下没有问题呢
但不在union下没有问题呢
东那个升 2010-02-25
- 打赏
- 举报
循环多，去重多。。。。。
BurningM 2010-02-25
- 打赏
- 举报
code上是有索引的，该加的索引都加了
with的那个表（产生树结构），数据量很小的，不会引起大问题
关键是partlife201002这个表，引起union后第2个查询非常慢
但是单独查询时一点问题都没有
with的那个表（产生树结构），数据量很小的，不会引起大问题
关键是partlife201002这个表，引起union后第2个查询非常慢
但是单独查询时一点问题都没有
starseeker7 2010-02-25
- 打赏
- 举报
而且基本上with就是後臺做個臨時表出來出來幫你處理問題，
你這樣複雜，然後結果集龐大的，，臨時表，，，又沒有做索引，，連接又多，，，自然問題就大了，，
推薦with查詢接到道臨時表裏，然後對臨時表做索引
再做你後續的查詢
你這樣複雜，然後結果集龐大的，，臨時表，，，又沒有做索引，，連接又多，，，自然問題就大了，，
推薦with查詢接到道臨時表裏，然後對臨時表做索引
再做你後續的查詢
starseeker7 2010-02-25
- 打赏
- 举报
o 代碼把腦殼看暈了。。。你也稍微縮進一下啊
留給老大們解決吧 - -
留給老大們解決吧 - -
feixianxxx 2010-02-25
- 打赏
- 举报
有没在code referencecount字段加索引
BurningM 2010-02-25
- 打赏
- 举报
就是第二个查询，没有referencecount的索引效率也很搞，但和第一个查询union后，没有这个索引效率大减
feixianxxx 2010-02-25
- 打赏
- 举报
你什么意思呢？
feixianxxx 2010-02-24
- 打赏
- 举报
鸭子 现在是 高手了
流氓兔 2010-02-24
- 打赏
- 举报
顶！
待测试
。。。。。。。。。。。。。。。。。。。。
待测试
。。。。。。。。。。。。。。。。。。。。
guguda2008 2010-02-24
- 打赏
- 举报
看一下计划
黄_瓜 2010-02-24
- 打赏
- 举报
不会，帮顶
BurningM 2010-02-24
- 打赏
- 举报
union all一样的
关键是union的只是两个数值而已，并没有多少结果集啊
关键是union的只是两个数值而已，并没有多少结果集啊
黄_瓜 2010-02-24
- 打赏
- 举报
1、得到的结果对吗？
2、如果是union惹的祸，那换成union all 如何？
2、如果是union惹的祸，那换成union all 如何？
ws_hgo 2010-02-24
- 打赏
- 举报
union
会消除重复的
所以耗时
会消除重复的
所以耗时
东那个升 2010-02-24
- 打赏
- 举报
相当于多了一步DISTINCT 运算
东那个升 2010-02-24
- 打赏
- 举报
因为你是UNION，所以当中有了去重的运算，所以耗时.
BurningM 2010-02-24
- 打赏
- 举报
第一个查询是
with PDMBOMCTE (TopCode, ParentCode, ChildCode, ChildCount, iLayer) as (select Code as TopCode, Code as ParentCode, Code as ChildCode, ChildCount = 1, iLayer = 1 from PDM_Part where Code = @Code union all select PDMBOMCTE.TopCode, PDM_BOM.ParentCode, PDM_Bom.ChildCode, PDM_Bom.ChildCount * PDMBOMCTE.ChildCount as ChildCount, iLayer +1 from PDM_Bom inner join PDMBOMCTE on PDM_BOM.ParentCode = PDMBOMCTE.ChildCode where not exists (select code from PDM_Part where (parttype = 5) and code = PDM_BOM.ParentCode))
select count(code) as 'Num' from PDM_Part where (PartType = 1 or PartType = 2 ) and exists (select * from PDMBOMCTE where PDMBOMCTE.Childcode = PDM_Part.Code)
第二个查询是
with PDMBOMCTE (TopCode, ParentCode, ChildCode, ChildCount, iLayer) as (select Code as TopCode, Code as ParentCode, Code as ChildCode, ChildCount = 1, iLayer = 1 from PDM_Part where Code = @Code union all select PDMBOMCTE.TopCode, PDM_BOM.ParentCode, PDM_Bom.ChildCode, PDM_Bom.ChildCount * PDMBOMCTE.ChildCount as ChildCount, iLayer +1 from PDM_Bom inner join PDMBOMCTE on PDM_BOM.ParentCode = PDMBOMCTE.ChildCode where not exists (select code from PDM_Part where (parttype = 5) and code = PDM_BOM.ParentCode))
select count(code) as 'Num' from PDM_Part where (PartType = 1 or PartType = 2 )
and exists (select * from sis_partlife201002 where producttype= '' and code = PDM_Part.code and (referencecount >= @over or (referencecount <= @lower and sumcount >= @ovsum))) and exists (select * from PDMBOMCTE where PDMBOMCTE.Childcode = PDM_Part.Code)
合起来就是
with PDMBOMCTE (TopCode, ParentCode, ChildCode, ChildCount, iLayer) as (select Code as TopCode, Code as ParentCode, Code as ChildCode, ChildCount = 1, iLayer = 1 from PDM_Part where Code = @Code union all select PDMBOMCTE.TopCode, PDM_BOM.ParentCode, PDM_Bom.ChildCode, PDM_Bom.ChildCount * PDMBOMCTE.ChildCount as ChildCount, iLayer +1 from PDM_Bom inner join PDMBOMCTE on PDM_BOM.ParentCode = PDMBOMCTE.ChildCode where not exists (select code from PDM_Part where (parttype = 5) and code = PDM_BOM.ParentCode))
select count(code) as 'Num' from PDM_Part where (PartType = 1 or PartType = 2 ) and exists (select * from PDMBOMCTE where PDMBOMCTE.Childcode = PDM_Part.Code)
union
select count(code) as 'Num' from PDM_Part where (PartType = 1 or PartType = 2 )
and exists (select * from sis_partlife201002 where producttype= '' and code = PDM_Part.code and (referencecount >= @over or (referencecount <= @lower and sumcount >= @ovsum))) and exists (select * from PDMBOMCTE where PDMBOMCTE.Childcode = PDM_Part.Code)
单独执行都很快，但union后就慢的可怕了，看计划似乎后面那个查询扫描过滤表的时间比较长，难道是(referencecount >= @over or (referencecount <= @lower and sumcount >= @ovsum)引起的？
其实这样的union我其他地方也有用的，但都没引起效率问题
with PDMBOMCTE (TopCode, ParentCode, ChildCode, ChildCount, iLayer) as (select Code as TopCode, Code as ParentCode, Code as ChildCode, ChildCount = 1, iLayer = 1 from PDM_Part where Code = @Code union all select PDMBOMCTE.TopCode, PDM_BOM.ParentCode, PDM_Bom.ChildCode, PDM_Bom.ChildCount * PDMBOMCTE.ChildCount as ChildCount, iLayer +1 from PDM_Bom inner join PDMBOMCTE on PDM_BOM.ParentCode = PDMBOMCTE.ChildCode where not exists (select code from PDM_Part where (parttype = 5) and code = PDM_BOM.ParentCode))
select count(code) as 'Num' from PDM_Part where (PartType = 1 or PartType = 2 ) and exists (select * from PDMBOMCTE where PDMBOMCTE.Childcode = PDM_Part.Code)
第二个查询是
with PDMBOMCTE (TopCode, ParentCode, ChildCode, ChildCount, iLayer) as (select Code as TopCode, Code as ParentCode, Code as ChildCode, ChildCount = 1, iLayer = 1 from PDM_Part where Code = @Code union all select PDMBOMCTE.TopCode, PDM_BOM.ParentCode, PDM_Bom.ChildCode, PDM_Bom.ChildCount * PDMBOMCTE.ChildCount as ChildCount, iLayer +1 from PDM_Bom inner join PDMBOMCTE on PDM_BOM.ParentCode = PDMBOMCTE.ChildCode where not exists (select code from PDM_Part where (parttype = 5) and code = PDM_BOM.ParentCode))
select count(code) as 'Num' from PDM_Part where (PartType = 1 or PartType = 2 )
and exists (select * from sis_partlife201002 where producttype= '' and code = PDM_Part.code and (referencecount >= @over or (referencecount <= @lower and sumcount >= @ovsum))) and exists (select * from PDMBOMCTE where PDMBOMCTE.Childcode = PDM_Part.Code)
合起来就是
with PDMBOMCTE (TopCode, ParentCode, ChildCode, ChildCount, iLayer) as (select Code as TopCode, Code as ParentCode, Code as ChildCode, ChildCount = 1, iLayer = 1 from PDM_Part where Code = @Code union all select PDMBOMCTE.TopCode, PDM_BOM.ParentCode, PDM_Bom.ChildCode, PDM_Bom.ChildCount * PDMBOMCTE.ChildCount as ChildCount, iLayer +1 from PDM_Bom inner join PDMBOMCTE on PDM_BOM.ParentCode = PDMBOMCTE.ChildCode where not exists (select code from PDM_Part where (parttype = 5) and code = PDM_BOM.ParentCode))
select count(code) as 'Num' from PDM_Part where (PartType = 1 or PartType = 2 ) and exists (select * from PDMBOMCTE where PDMBOMCTE.Childcode = PDM_Part.Code)
union
select count(code) as 'Num' from PDM_Part where (PartType = 1 or PartType = 2 )
and exists (select * from sis_partlife201002 where producttype= '' and code = PDM_Part.code and (referencecount >= @over or (referencecount <= @lower and sumcount >= @ovsum))) and exists (select * from PDMBOMCTE where PDMBOMCTE.Childcode = PDM_Part.Code)
单独执行都很快，但union后就慢的可怕了，看计划似乎后面那个查询扫描过滤表的时间比较长，难道是(referencecount >= @over or (referencecount <= @lower and sumcount >= @ovsum)引起的？
其实这样的union我其他地方也有用的，但都没引起效率问题
东那个升 2010-02-24
- 打赏
- 举报
最好把你的这串代码贴出来看看
