使用多个 OUTER APPLY 优化查询

Posted 2023-04-15

【中文标题】使用多个 OUTER APPLY 优化查询

【英文标题】：Optimize query with multiple OUTER APPLY

【发布时间】：2018-03-08 11:12:32

【问题描述】：

我正在查询多个OUTER APPLY，但所有表在连接列上都有主键（因此此处使用聚集索引），所以我不知道如何进一步优化此查询。此外，这里也不可能使用索引视图，因为它们禁止使用ORDER BY 和TOP。

所以我有

表 Fields 带有 Id 主键和大量其他列。

WeatherHistory 具有复杂主键（FieldId 和 [Date]）和很多列的表，

NdviImageHistory 表与FieldId、[Date]、[Base64] 列（复杂主键FieldId 和[Date]）其中[Base64] 存储图像base64，

NaturalColorImageHistory 表与FieldId、[Date]、[Base64] 列（复杂主键FieldId 和[Date]）其中[Base64] 存储图像base64，

NdviHistory 表与 FieldId、[Date]、MeanNdvi 列（复杂主键 FieldId 和 [Date]），

FieldSeasonHistory 表与 Field、StartDate、EndDate 列（复杂主键 FieldId 和 [Date]）。

我的查询

SELECT Fields.*,
    WeatherHistory.TempSumC AS CurrentTempSumC,
    TempSumF AS CurrentTempSumF,
    PrecipitationSumMm AS CurrentPrecipitationSumMm,
    nih.[Base64] AS CurrentNdviImageBase64,
    ncih.[Base64] AS CurrentNaturalColorImageBase64,
    MeanNdvi AS CurrentMeanNdvi,
    IsOpenSeason
FROM Fields
LEFT JOIN WeatherHistory ON FieldId = Id AND [Date] = CAST(GETUTCDATE() AS DATE)
OUTER APPLY
(
    SELECT TOP 1 [Base64]
    FROM NdviImageHistory
    WHERE FieldId = Id
    ORDER BY [Date] DESC
) nih
OUTER APPLY
(
    SELECT TOP 1 [Base64]
    FROM NaturalColorImageHistory
    WHERE FieldId = Id
    ORDER BY [Date] DESC
) ncih
OUTER APPLY
(
    SELECT TOP 1 MeanNdvi
    FROM NdviHistory
    WHERE FieldId = Id
    ORDER BY [Date] DESC
) nh
OUTER APPLY
(
    SELECT TOP 1 CASE WHEN EndDate IS NULL THEN 1 ELSE 0 END AS IsOpenSeason
    FROM FieldSeasonHistory
    WHERE FieldId = Id
    ORDER BY [StartDate] DESC
) fsh
WHERE UserId = (SELECT Id FROM Users WHERE Email = @email) AND IsArchived = 0

我没有创建任何索引，因为我认为自动生成的集群索引（基于主键）应该足够了（但我可能是错的）。此查询执行大约 15 秒，但我想减少查询时间。

---

编辑：为Fields 表的UserId 和IsArchived 列添加了索引。查询执行计划：

---

** 编辑 2：** 统计：

SQL Server parse and compile time: 
   CPU time = 0 ms, elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.
SQL Server parse and compile time: 
   CPU time = 0 ms, elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

(13 row(s) affected)
Table 'FieldSeasonHistory'. Scan count 13, logical reads 26, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'NdviHistory'. Scan count 13, logical reads 26, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'NaturalColorImageHistory'. Scan count 13, logical reads 26, physical reads 0, read-ahead reads 0, lob logical reads 604, lob physical reads 0, lob read-ahead reads 0.
Table 'NdviImageHistory'. Scan count 13, logical reads 39, physical reads 0, read-ahead reads 0, lob logical reads 68, lob physical reads 0, lob read-ahead reads 0.
Table 'WeatherHistory'. Scan count 0, logical reads 39, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'Users'. Scan count 0, logical reads 228, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
Table 'Fields'. Scan count 1, logical reads 19, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

(1 row(s) affected)

 SQL Server Execution Times:
   CPU time = 15 ms,  elapsed time = 16 ms.
SQL Server parse and compile time: 
   CPU time = 0 ms, elapsed time = 0 ms.

 SQL Server Execution Times:
   CPU time = 0 ms,  elapsed time = 0 ms.

【参考方案1】：

每个表都需要多列索引。索引应该是 where 中的列，order by 中的列，然后是 select 中的列。例如：

NdviImageHistory(FieldId, [Date], [Base64]) NaturalColorImageHistory(FieldId, [Date], [Base64]) 等等。

【讨论】：

我需要为每一列创建单独的索引还是为多列创建单个索引？

我的意思是每表一个还是每列一个？

我也无法索引[Base64]，因为它是VARCHAR(MAX) 列。

@VadimOvchinnikov 。 . .然后使用前两列创建索引。这些是复合索引——具有多列的索引。

好的，我明白了。对于表NdviImageHistory 和NaturalColorImageHistory，已经有聚集索引，因为前两列是复合主键。我还假设问题与这些表有关，因为如果我从查询中删除它们，查询会在一秒钟内执行。

【参考方案2】：

像这样更新您的查询。

SELECT Fields.*,
    WeatherHistory.TempSumC AS CurrentTempSumC,
    TempSumF AS CurrentTempSumF,
    PrecipitationSumMm AS CurrentPrecipitationSumMm,
    nih.[Base64] AS CurrentNdviImageBase64,
    ncih.[Base64] AS CurrentNaturalColorImageBase64,
    MeanNdvi AS CurrentMeanNdvi,
    IsOpenSeason
FROM Fields
LEFT JOIN WeatherHistory ON FieldId = Id AND [Date] = CAST(GETUTCDATE() AS DATE)
OUTER APPLY
(
    SELECT MAX(NdviImageHistory.ID) MAX_ID
    FROM NdviImageHistory
    WHERE FieldId = Id

) nih_ID
OUTER APPLY
(
    SELECT [Base64] FROM NdviImageHistory X WHERE X.ID = nih_ID.MAX_ID
)nih

用户更多外申请所有这些表 nih , ncih, nh, fsh 试试这个。

我只为 [nih] 使用了 1 个外部

从 OUTER APPLY 连接中删除 TOP 1 和 Order by

【讨论】：

你确定这会编译吗？

是的，它的工作原理，如果你可以为我在解决方案中提到的每个表添加更多外部。

好的，为什么你认为这会比原始查询执行得更快？