ROW_NUMBER() 大查询超时。最佳解决方案？
ROW_NUMBER() OVER timing out with large query. Best solution?
我正在尝试修复应用程序（不是我自己的应用程序）中的一个错误，该错误 运行 是对数据库的大型查询。这会在部署时导致 504 网关超时。当我 运行 MSSMS 中的查询时，它会永远执行而不会返回任何结果：
SELECT *
FROM (SELECT Row_number()
OVER (
ORDER BY [TABLE_ONE].[MET_uid] )
AS RowNum,
(SELECT Count(1)
FROM TABLE_ONE
LEFT OUTER JOIN TABLE_TWO
ON TABLE_ONE.SCC_uid =
TABLE_TWO.SCC_uid
WHERE ( [TABLE_ONE].[ANOTHER_DATE] <=
'30 November 2020' )
AND ( [TABLE_ONE].[SOME_DATE] >=
'30 November 2020' )
AND ( [TABLE_TWO].[ANOTHER_DATE] <=
'30 November 2020' )
AND ( [TABLE_TWO].[SOME_DATE] >=
'30 November 2020' ))
AS
TotalRows,
[TABLE_ONE].[MET_uid]
AS TABLE_ONE_MET_UID,
[TABLE_ONE].[MET_type_uid]
AS TABLE_ONE_MET_TYPE_UID,
[TABLE_ONE].[IIC]
AS
TABLE_ONE_IIC,
[TABLE_ONE].[LCD]
AS
TABLE_ONE_LCD,
[TABLE_ONE].[ANOTHER_DATE]
AS TABLE_ONE_ANOTHER_DATE,
[TABLE_ONE].[SOME_DATE]
AS TABLE_ONE_SOME_DATE,
[TABLE_ONE].[SOME_DATEd]
AS TABLE_ONE_SOME_DATED,
[TABLE_ONE].[LED]
AS TABLE_ONE_LED,
[TABLE_ONE].[LED_U_UID]
AS
TABLE_ONE_LED_U_UID,
[TABLE_ONE].[P_UID]
AS TABLE_ONE_P_UID,
[TABLE_ONE].[FM_UID]
AS TABLE_ONE_FM_UID,
[TABLE_ONE].[D_LEN]
AS TABLE_ONE_D_LEN,
[TABLE_ONE].[D_AR]
AS TABLE_ONE_D_AR,
[TABLE_ONE].[ETI]
AS TABLE_ONE_ETI,
[TABLE_ONE].[NT]
AS TABLE_ONE_NT,
[TABLE_ONE].[BLT]
AS TABLE_ONE_BLT,
[TABLE_ONE].[BLK]
AS TABLE_ONE_BLK,
[TABLE_ONE].[BLU]
AS TABLE_ONE_BLU,
[TABLE_ONE].[trcn]
AS TABLE_ONE_TRCN,
[TABLE_ONE].[SCC_uid]
AS TABLE_ONE_SCC_UID,
[TABLE_TWO].[F_uid]
AS TABLE_TWO_F_UID,
[TABLE_TWO].[RDT_UID]
AS TABLE_TWO_RDT_UID,
[TABLE_TWO].[SCC_code]
AS TABLE_TWO_SCC_CODE,
[TABLE_TWO].[SCC_name]
AS TABLE_TWO_SCC_NAME,
[TABLE_TWO].[SCC_type_uid]
AS TABLE_TWO_SCC_TYPE_UID,
[TABLE_TWO].[SCC_length]
AS TABLE_TWO_SCC_LENGTH,
[TABLE_TWO].[SPDT]
AS TABLE_TWO_SPDT,
[TABLE_TWO].[SNC]
AS TABLE_TWO_SNC,
[TABLE_TWO].[SNJT_uid]
AS
TABLE_TWO_SNJT_UID,
[TABLE_TWO].[SN_name]
AS TABLE_TWO_SN_NAME,
[TABLE_TWO].[ENC]
AS TABLE_TWO_ENC,
[TABLE_TWO].[ENCT_JT_uid]
AS
TABLE_TWO_ENCT_JT_UID,
[TABLE_TWO].[ENCT_name]
AS TABLE_TWO_ENCT_NAME,
[TABLE_TWO].[NTT_uid]
AS TABLE_TWO_NTT_UID,
[TABLE_TWO].[RG_uid]
AS TABLE_TWO_RG_UID,
[TABLE_TWO].[AR_uid]
AS TABLE_TWO_AR_UID,
[TABLE_TWO].[division_uid]
AS TABLE_TWO_DIVISION_UID,
[TABLE_TWO].[DST_uid]
AS TABLE_TWO_DST_UID,
[TABLE_TWO].[SVV_AR_uid]
AS TABLE_TWO_SVV_AR_UID,
[TABLE_TWO].[RTF]
AS TABLE_TWO_RTF,
[TABLE_TWO].[CL_UID]
AS TABLE_TWO_CL_UID,
[TABLE_TWO].[MH_uid]
AS
TABLE_TWO_MH_UID,
[TABLE_TWO].[ENV_uid]
AS TABLE_TWO_ENV_UID,
[TABLE_TWO].[ONR_uid]
AS TABLE_TWO_ONR_UID,
[TABLE_TWO].[AG_uid]
AS TABLE_TWO_AG_UID,
[TABLE_TWO].[ANOTHER_DATE]
AS TABLE_TWO_ANOTHER_DATE,
[TABLE_TWO].[SOME_DATE]
AS TABLE_TWO_SOME_DATE,
[TABLE_TWO].[DPT_code]
AS TABLE_TWO_DPT_CODE,
[TABLE_TWO].[NMT_code]
AS TABLE_TWO_NMT_CODE,
[TABLE_TWO].[RW_code]
AS
TABLE_TWO_RW_CODE,
[TABLE_TWO].[SLVL_uid]
AS TABLE_TWO_SLVL_UID,
[TABLE_TWO].[FTG_code]
AS TABLE_TWO_FTG_CODE,
[TABLE_TWO].[FT_uid]
AS
TABLE_TWO_FT_UID,
[TABLE_TWO].[FO_uid]
AS TABLE_TWO_FO_UID,
[TABLE_TWO].[DC]
AS TABLE_TWO_DC,
[TABLE_TWO].[SVV_uid]
AS TABLE_TWO_SVV_UID,
[TABLE_TWO].[SCC_length_status]
AS
TABLE_TWO_SCC_LENGTH_STATUS,
[TABLE_TWO].[SVV]
AS TABLE_TWO_SVV,
[TABLE_TWO].[MAX]
AS TABLE_TWO_MAX,
[TABLE_TWO].[SQN]
AS TABLE_TWO_SQN,
[TABLE_TWO].[DRT]
AS TABLE_TWO_DRT,
[TABLE_TWO].[cwxsp]
AS TABLE_TWO_CWXSP,
[TABLE_TWO].[offcwxsp]
AS TABLE_TWO_OFFCWXSP,
[TABLE_TWO].[ukpms_MHH_uid]
AS
TABLE_TWO_UKPMS_MHH_UID,
[TABLE_TWO].[fway_MHH_uid]
AS
TABLE_TWO_FWAY_MHH_UID,
[TABLE_TWO].[SMH_uid]
AS
TABLE_TWO_SMH_UID,
[TABLE_TWO].[SMP_ROU_uid]
AS
TABLE_TWO_SMP_ROU_UID,
[TABLE_TWO].[SMP_dot_CLK_uid]
AS
TABLE_TWO_SMP_DOT_CLK_UID,
[TABLE_TWO].[LCT_uid]
AS TABLE_TWO_LCT_UID,
[TABLE_TWO].[LH]
AS TABLE_TWO_LH,
[TABLE_TWO].[programmed_date]
AS TABLE_TWO_PROGRAMMED_DATE,
[TABLE_TWO].[LCD_uid]
AS TABLE_TWO_LCD_UID,
[TABLE_TWO].[CW]
AS TABLE_TWO_CW,
[TABLE_TWO].[lca_occ_LCT_uid]
AS
TABLE_TWO_LCA_OCC_LCT_UID,
[TABLE_TWO].[SG]
AS TABLE_TWO_SG,
[TABLE_TWO].[SGG]
AS TABLE_TWO_SGG
FROM TABLE_ONE
LEFT OUTER JOIN TABLE_TWO
ON TABLE_ONE.SCC_uid =
TABLE_TWO.SCC_uid
WHERE ( [TABLE_ONE].[ANOTHER_DATE] <= '30 November 2020' )
AND ( [TABLE_ONE].[SOME_DATE] >= '30 November 2020' )
AND ( [TABLE_TWO].[ANOTHER_DATE] <= '30 November 2020' )
AND ( [TABLE_TWO].[SOME_DATE] >= '30 November 2020' )) AS
RowConstrainedResult
WHERE rownum >= 0
AND rownum < 24
ORDER BY rownum
我在不使用 ROW_NUMBERS() OVER 的情况下简化了查询，查询执行得相当快，没有问题。这是修改后的查询：
SELECT dbo.TABLE_ONE.*,
dbo.TABLE_TWO.*
FROM dbo.TABLE_ONE
INNER JOIN dbo.TABLE_TWO
ON dbo.TABLE_ONE.SCC_uid =
dbo.TABLE_TWO.SCC_uid
WHERE ( dbo.TABLE_ONE.ANOTHER_DATE <=
CONVERT(DATETIME, '2020-11-30 00:00:00'
,
102) )
AND ( dbo.TABLE_ONE.SOME_DATE >=
CONVERT(DATETIME, '2020-11-30 00:00:00',
102) )
AND ( dbo.TABLE_TWO.ANOTHER_DATE <=
CONVERT(DATETIME, '2020-11-30 00:00:00',
102) )
AND ( dbo.TABLE_TWO.SOME_DATE >=
CONVERT(DATETIME, '2020-11-30 00:00:00',
102) )
在应用程序中，在它超时之前它似乎正在检索 160 万条记录，而简化查询只有 72k，不知道为什么会这样。
知道如何解决这个问题吗？
首先，将查询表述为 CTE：
WITH cs as (
SELECT ca.*, sa.*. -- you may need to adjust the columns so there are no duplicates
FROM dbo.RM_COMMON_ATTRIBUTES ca INNER JOIN
dbo.SECTION_ATTRIBUTES sa
ON ca.SECTION_UID = sa.SECTION_UID
WHERE ca.START_DATE <= '2020-11-30' AND
ca.END_DATE >= '2020-11-30' AND
sa.START_DATE <= '2020-11-30' AND
sa.END_DATE >= '2020-11-30'
)
然后，有几件事可能会影响性能：
- 结果集的大小。
- 行号。
- 返回 所有 行，而不仅仅是第一行。
要排除第三种可能性，只需检查它的运行速度：
select count(*)
from cs;
您可以通过简单地执行以下操作来检查第一个：
select *
from cs;
然后可以在row_number()中添加：
SELECT cs.*,
ROW_NUMBER OVER ( ORDER BY [ITEM_UID] ) AS RowNum
FROM cs;
我正在尝试修复应用程序（不是我自己的应用程序）中的一个错误，该错误 运行 是对数据库的大型查询。这会在部署时导致 504 网关超时。当我 运行 MSSMS 中的查询时，它会永远执行而不会返回任何结果：
SELECT *
FROM (SELECT Row_number()
OVER (
ORDER BY [TABLE_ONE].[MET_uid] )
AS RowNum,
(SELECT Count(1)
FROM TABLE_ONE
LEFT OUTER JOIN TABLE_TWO
ON TABLE_ONE.SCC_uid =
TABLE_TWO.SCC_uid
WHERE ( [TABLE_ONE].[ANOTHER_DATE] <=
'30 November 2020' )
AND ( [TABLE_ONE].[SOME_DATE] >=
'30 November 2020' )
AND ( [TABLE_TWO].[ANOTHER_DATE] <=
'30 November 2020' )
AND ( [TABLE_TWO].[SOME_DATE] >=
'30 November 2020' ))
AS
TotalRows,
[TABLE_ONE].[MET_uid]
AS TABLE_ONE_MET_UID,
[TABLE_ONE].[MET_type_uid]
AS TABLE_ONE_MET_TYPE_UID,
[TABLE_ONE].[IIC]
AS
TABLE_ONE_IIC,
[TABLE_ONE].[LCD]
AS
TABLE_ONE_LCD,
[TABLE_ONE].[ANOTHER_DATE]
AS TABLE_ONE_ANOTHER_DATE,
[TABLE_ONE].[SOME_DATE]
AS TABLE_ONE_SOME_DATE,
[TABLE_ONE].[SOME_DATEd]
AS TABLE_ONE_SOME_DATED,
[TABLE_ONE].[LED]
AS TABLE_ONE_LED,
[TABLE_ONE].[LED_U_UID]
AS
TABLE_ONE_LED_U_UID,
[TABLE_ONE].[P_UID]
AS TABLE_ONE_P_UID,
[TABLE_ONE].[FM_UID]
AS TABLE_ONE_FM_UID,
[TABLE_ONE].[D_LEN]
AS TABLE_ONE_D_LEN,
[TABLE_ONE].[D_AR]
AS TABLE_ONE_D_AR,
[TABLE_ONE].[ETI]
AS TABLE_ONE_ETI,
[TABLE_ONE].[NT]
AS TABLE_ONE_NT,
[TABLE_ONE].[BLT]
AS TABLE_ONE_BLT,
[TABLE_ONE].[BLK]
AS TABLE_ONE_BLK,
[TABLE_ONE].[BLU]
AS TABLE_ONE_BLU,
[TABLE_ONE].[trcn]
AS TABLE_ONE_TRCN,
[TABLE_ONE].[SCC_uid]
AS TABLE_ONE_SCC_UID,
[TABLE_TWO].[F_uid]
AS TABLE_TWO_F_UID,
[TABLE_TWO].[RDT_UID]
AS TABLE_TWO_RDT_UID,
[TABLE_TWO].[SCC_code]
AS TABLE_TWO_SCC_CODE,
[TABLE_TWO].[SCC_name]
AS TABLE_TWO_SCC_NAME,
[TABLE_TWO].[SCC_type_uid]
AS TABLE_TWO_SCC_TYPE_UID,
[TABLE_TWO].[SCC_length]
AS TABLE_TWO_SCC_LENGTH,
[TABLE_TWO].[SPDT]
AS TABLE_TWO_SPDT,
[TABLE_TWO].[SNC]
AS TABLE_TWO_SNC,
[TABLE_TWO].[SNJT_uid]
AS
TABLE_TWO_SNJT_UID,
[TABLE_TWO].[SN_name]
AS TABLE_TWO_SN_NAME,
[TABLE_TWO].[ENC]
AS TABLE_TWO_ENC,
[TABLE_TWO].[ENCT_JT_uid]
AS
TABLE_TWO_ENCT_JT_UID,
[TABLE_TWO].[ENCT_name]
AS TABLE_TWO_ENCT_NAME,
[TABLE_TWO].[NTT_uid]
AS TABLE_TWO_NTT_UID,
[TABLE_TWO].[RG_uid]
AS TABLE_TWO_RG_UID,
[TABLE_TWO].[AR_uid]
AS TABLE_TWO_AR_UID,
[TABLE_TWO].[division_uid]
AS TABLE_TWO_DIVISION_UID,
[TABLE_TWO].[DST_uid]
AS TABLE_TWO_DST_UID,
[TABLE_TWO].[SVV_AR_uid]
AS TABLE_TWO_SVV_AR_UID,
[TABLE_TWO].[RTF]
AS TABLE_TWO_RTF,
[TABLE_TWO].[CL_UID]
AS TABLE_TWO_CL_UID,
[TABLE_TWO].[MH_uid]
AS
TABLE_TWO_MH_UID,
[TABLE_TWO].[ENV_uid]
AS TABLE_TWO_ENV_UID,
[TABLE_TWO].[ONR_uid]
AS TABLE_TWO_ONR_UID,
[TABLE_TWO].[AG_uid]
AS TABLE_TWO_AG_UID,
[TABLE_TWO].[ANOTHER_DATE]
AS TABLE_TWO_ANOTHER_DATE,
[TABLE_TWO].[SOME_DATE]
AS TABLE_TWO_SOME_DATE,
[TABLE_TWO].[DPT_code]
AS TABLE_TWO_DPT_CODE,
[TABLE_TWO].[NMT_code]
AS TABLE_TWO_NMT_CODE,
[TABLE_TWO].[RW_code]
AS
TABLE_TWO_RW_CODE,
[TABLE_TWO].[SLVL_uid]
AS TABLE_TWO_SLVL_UID,
[TABLE_TWO].[FTG_code]
AS TABLE_TWO_FTG_CODE,
[TABLE_TWO].[FT_uid]
AS
TABLE_TWO_FT_UID,
[TABLE_TWO].[FO_uid]
AS TABLE_TWO_FO_UID,
[TABLE_TWO].[DC]
AS TABLE_TWO_DC,
[TABLE_TWO].[SVV_uid]
AS TABLE_TWO_SVV_UID,
[TABLE_TWO].[SCC_length_status]
AS
TABLE_TWO_SCC_LENGTH_STATUS,
[TABLE_TWO].[SVV]
AS TABLE_TWO_SVV,
[TABLE_TWO].[MAX]
AS TABLE_TWO_MAX,
[TABLE_TWO].[SQN]
AS TABLE_TWO_SQN,
[TABLE_TWO].[DRT]
AS TABLE_TWO_DRT,
[TABLE_TWO].[cwxsp]
AS TABLE_TWO_CWXSP,
[TABLE_TWO].[offcwxsp]
AS TABLE_TWO_OFFCWXSP,
[TABLE_TWO].[ukpms_MHH_uid]
AS
TABLE_TWO_UKPMS_MHH_UID,
[TABLE_TWO].[fway_MHH_uid]
AS
TABLE_TWO_FWAY_MHH_UID,
[TABLE_TWO].[SMH_uid]
AS
TABLE_TWO_SMH_UID,
[TABLE_TWO].[SMP_ROU_uid]
AS
TABLE_TWO_SMP_ROU_UID,
[TABLE_TWO].[SMP_dot_CLK_uid]
AS
TABLE_TWO_SMP_DOT_CLK_UID,
[TABLE_TWO].[LCT_uid]
AS TABLE_TWO_LCT_UID,
[TABLE_TWO].[LH]
AS TABLE_TWO_LH,
[TABLE_TWO].[programmed_date]
AS TABLE_TWO_PROGRAMMED_DATE,
[TABLE_TWO].[LCD_uid]
AS TABLE_TWO_LCD_UID,
[TABLE_TWO].[CW]
AS TABLE_TWO_CW,
[TABLE_TWO].[lca_occ_LCT_uid]
AS
TABLE_TWO_LCA_OCC_LCT_UID,
[TABLE_TWO].[SG]
AS TABLE_TWO_SG,
[TABLE_TWO].[SGG]
AS TABLE_TWO_SGG
FROM TABLE_ONE
LEFT OUTER JOIN TABLE_TWO
ON TABLE_ONE.SCC_uid =
TABLE_TWO.SCC_uid
WHERE ( [TABLE_ONE].[ANOTHER_DATE] <= '30 November 2020' )
AND ( [TABLE_ONE].[SOME_DATE] >= '30 November 2020' )
AND ( [TABLE_TWO].[ANOTHER_DATE] <= '30 November 2020' )
AND ( [TABLE_TWO].[SOME_DATE] >= '30 November 2020' )) AS
RowConstrainedResult
WHERE rownum >= 0
AND rownum < 24
ORDER BY rownum
我在不使用 ROW_NUMBERS() OVER 的情况下简化了查询，查询执行得相当快，没有问题。这是修改后的查询：
SELECT dbo.TABLE_ONE.*,
dbo.TABLE_TWO.*
FROM dbo.TABLE_ONE
INNER JOIN dbo.TABLE_TWO
ON dbo.TABLE_ONE.SCC_uid =
dbo.TABLE_TWO.SCC_uid
WHERE ( dbo.TABLE_ONE.ANOTHER_DATE <=
CONVERT(DATETIME, '2020-11-30 00:00:00'
,
102) )
AND ( dbo.TABLE_ONE.SOME_DATE >=
CONVERT(DATETIME, '2020-11-30 00:00:00',
102) )
AND ( dbo.TABLE_TWO.ANOTHER_DATE <=
CONVERT(DATETIME, '2020-11-30 00:00:00',
102) )
AND ( dbo.TABLE_TWO.SOME_DATE >=
CONVERT(DATETIME, '2020-11-30 00:00:00',
102) )
在应用程序中，在它超时之前它似乎正在检索 160 万条记录，而简化查询只有 72k，不知道为什么会这样。
知道如何解决这个问题吗？
首先，将查询表述为 CTE：
WITH cs as (
SELECT ca.*, sa.*. -- you may need to adjust the columns so there are no duplicates
FROM dbo.RM_COMMON_ATTRIBUTES ca INNER JOIN
dbo.SECTION_ATTRIBUTES sa
ON ca.SECTION_UID = sa.SECTION_UID
WHERE ca.START_DATE <= '2020-11-30' AND
ca.END_DATE >= '2020-11-30' AND
sa.START_DATE <= '2020-11-30' AND
sa.END_DATE >= '2020-11-30'
)
然后，有几件事可能会影响性能：
- 结果集的大小。
- 行号。
- 返回 所有 行，而不仅仅是第一行。
要排除第三种可能性，只需检查它的运行速度：
select count(*)
from cs;
您可以通过简单地执行以下操作来检查第一个：
select *
from cs;
然后可以在row_number()中添加：
SELECT cs.*,
ROW_NUMBER OVER ( ORDER BY [ITEM_UID] ) AS RowNum
FROM cs;