如何最好地处理这个执行计划
How to best approach this execution plan
我们的仓库管理包存在存储过程瓶颈(众多瓶颈之一),主要的减速是由于生成此执行计划的查询。
https://www.brentozar.com/pastetheplan/?id=HkNg65elP
存储过程可能需要 3 到 10 秒到 运行,这在其 运行 所在的业务流程上下文中相当慢。
一些附加信息:是的,有一个 table 正在执行完整的 table 扫描,但是这个 table 很窄,只有 76 行。查询进行了一些左连接和一些排序,这是产生正确的最高结果所必需的。总的来说,它有点像“Rube Goldberg”类型的查询,可能会被简化,但我的目标是看看是否有可能帮助建立一些索引(我已经完成了,它有点帮助)甚至一些如果需要,对查询进行小的调整。
最后,我需要根据计划知道接下来的重点。
这里是查询:
SELECT TOP 1 loc.location_id, loc.wh_id
FROM t_item_uom itu WITH (NOLOCK)
INNER JOIN t_class_loca clc WITH (NOLOCK)
ON itu.wh_id = clc.wh_id
AND ISNULL(dbo.usf_get_item_class_dia_ovrd ('13098271', '895', itu.uom, NULL), itu.class_id) = clc.class_id
INNER JOIN t_location loc WITH (NOLOCK)
ON clc.wh_id = loc.wh_id
AND clc.location_id = loc.location_id
INNER JOIN t_pick_area pka WITH (NOLOCK)
ON pka.pick_area = loc.pick_area
AND pka.wh_id = loc.wh_id
AND (pka.pick_area <> N'LABEL' OR (pka.pick_area = N'LABEL' AND 0 IS NULL AND 0 IS NULL) )
AND (pka.pick_area_type = 'R' OR (pka.pick_area_type = 'V' and 0 IS NULL) )
INNER JOIN t_zone_loca zlc WITH (NOLOCK)
ON loc.wh_id = zlc.wh_id
AND loc.location_id = zlc.location_id
INNER JOIN (
SELECT loc.wh_id, loc.pnd_location_id --, loc.location_id
FROM t_location loc with (nolock)
inner join t_class_loca clc WITH (NOLOCK)
on clc.location_id = loc.location_id
and clc.wh_id = loc.wh_id
and clc.class_id = 'APPAREL'
LEFT JOIN t_stored_item sto with (nolock)
ON sto.put_away_location = loc.location_id
AND sto.wh_id = loc.wh_id --BTH 20160907 missing wh_id
AND sto.put_away_location IS NOT NULL
AND sto.type = 0
WHERE loc.type in ('I','M')
AND loc.pnd_location_id IS NOT NULL --BTH 20160907 - remove from having clause, add here
GROUP BY loc.wh_id, loc.pnd_location_id, loc.c3
HAVING ((COUNT(sto.hu_id) < 100
and loc.pnd_location_id IS NOT NULL --BTH 201600907
and c3 is null)
OR (COUNT(sto.hu_id) < 500 --and loc.pnd_location_id IS NOT NULL --BTH 201600907
and c3 = 'BULK'))
) as pnd
ON pnd.wh_id = loc.wh_id
AND pnd.pnd_location_id = loc.pnd_location_id
LEFT OUTER JOIN t_put_rules_empty_and_unalloc_locs_by_pnd tpr WITH (NOLOCK)
ON tpr.pnd_location_id = loc.pnd_location_id
AND tpr.class_id = itu.class_id
AND tpr.wh_id = loc.wh_id
LEFT OUTER JOIN t_work_q q WITH(NOLOCK)
ON q.location_id = loc.location_id
AND q.wh_id = loc.wh_id
AND q.work_type = '08'
AND q.work_status = 'U'
WHERE loc.status = 'E'
AND ISNULL(q.work_q_id, 0) = 0
AND ( loc.c3 is null or loc.c3 not in ('R','H','S'))
AND (
(
loc.type = 'M'
AND ( (SELECT TOP 1 max_sku_count
FROM t_zone zone2 (NOLOCK)
WHERE zone2.wh_id = loc.wh_id
AND loc.zone = zone2.zone) >
(SELECT COUNT(sto2.item_number)
FROM t_stored_item sto2 (NOLOCK)
WHERE loc.wh_id = sto2.wh_id
AND loc.location_id = sto2.location_id
)
OR '13098271' IN (SELECT sto2.item_number
FROM t_stored_item sto2 (NOLOCK)
WHERE loc.wh_id = sto2.wh_id
AND loc.location_id = sto2.location_id
)
)
)
OR (loc.type = 'I' AND itu.unit_volume = 0 AND itu.nested_volume = 0)
OR (loc.type = 'I' AND loc.capacity_volume = 0)
OR (loc.type = 'I' AND loc.capacity_volume >= 0 +
(CASE
WHEN 0 = 0 THEN 0
ELSE 0
END * (1 - 1)
)
)
)
--Ensure that only one item is designated to the location
AND (loc.type = 'M'
OR (loc.type = 'I' AND NOT EXISTS (SELECT 1
FROM t_stored_item sto2 WITH (NOLOCK) -- Sum the item in the location to determine volume
WHERE sto2.wh_id = loc.wh_id
AND sto2.put_away_location = loc.location_id)))
AND itu.wh_id = '895'
AND itu.item_number = '13098271'
AND clc.class_id = 'APPAREL'
AND zlc.zone = 'ALL'
ORDER BY
tpr.percent_empty_and_unalloc DESC,
loc.type,
loc.user_count,
loc.picking_flow,
loc.location_id
在不知道您的表和数据的结构的情况下,我的建议是查看执行成本较高的部分。在您的示例中,这将位于右上角:分别为 14%(内部连接)、17% 和 22%,+ 19% 和 25% 其他地方。
其他重要的事情:您的 索引 以及它们是否按应有的方式使用。我觉得不是。
关注 25%(Key Lookup (Clustered)):这 link 可能会帮助您更好地理解问题(并省去冗长的解释)。同样,我不知道你的表的结构。但是我觉得你的索引在这里不够。
我看到了这个:CONVERT_IMPLICIT(nvarchar(4000)
。这是什么 ?会不会降低性能?
如果表的结构不好并且数据模型错误,那么优化起来会更加困难。添加更多索引或改写查询并不总是解决方案。
基于快速浏览您的执行计划。根据成本,第 1 期和第 3 期将为您提供最佳投资回报率。
第一期:Key Lookup
在您的索引 IDX_wh_id_status_pnd_location_id
中,您的索引的 INCLUDE 部分缺少列 zone
。
简而言之:您正在比较不同类型的列。确保比较相同类型的列。如果这些是外键列,则两个表中的类型应该完全相同。如果它们是参数,请更改类型或 cast/covert 它们。
第三期:聚合
您在 [t_stored_item].[sto].put_away_location
上有一个聚合(最大值、计数、平均值...),行估计值为 129108。
尝试创建一个 indexed view for this part of the query with the aggregation in the indexed view. Use the indexed view instead. More info
估计也与实际相差甚远,您可以尝试重建统计数据,但可能无济于事。为什么? Read this
第四期:用户自定义函数
您有一个 INNER JOIN usf_get_item_class_dia_ovrd
是否可以内联编写用户定义函数的逻辑?当我们在内联时,代码通常会被优化,现在标量函数正在逐行执行而不是基于集合。
第五期:常量扫描-实际行数0
这可能不是什么大问题,但当表达式相互抵消时,通常会发生这种情况。虚拟示例:1 = 0
将始终评估为 0 行,因此 SQL 服务器将其替换为空常量扫描。
在复杂的查询中,您可能无法立即找到它。这不会对性能产生很大影响,但是当您从查询中删除这些时,您可能会获得更好的执行计划。
如果有兴趣观看 this video 以更好地了解查询优化器。 (有点旧,但仍然相关)
奖励:参数嗅探
您提到它是一个存储过程。存储过程通常有 parameters sniffing.
的问题
我们的仓库管理包存在存储过程瓶颈(众多瓶颈之一),主要的减速是由于生成此执行计划的查询。
https://www.brentozar.com/pastetheplan/?id=HkNg65elP
存储过程可能需要 3 到 10 秒到 运行,这在其 运行 所在的业务流程上下文中相当慢。
一些附加信息:是的,有一个 table 正在执行完整的 table 扫描,但是这个 table 很窄,只有 76 行。查询进行了一些左连接和一些排序,这是产生正确的最高结果所必需的。总的来说,它有点像“Rube Goldberg”类型的查询,可能会被简化,但我的目标是看看是否有可能帮助建立一些索引(我已经完成了,它有点帮助)甚至一些如果需要,对查询进行小的调整。
最后,我需要根据计划知道接下来的重点。
这里是查询:
SELECT TOP 1 loc.location_id, loc.wh_id
FROM t_item_uom itu WITH (NOLOCK)
INNER JOIN t_class_loca clc WITH (NOLOCK)
ON itu.wh_id = clc.wh_id
AND ISNULL(dbo.usf_get_item_class_dia_ovrd ('13098271', '895', itu.uom, NULL), itu.class_id) = clc.class_id
INNER JOIN t_location loc WITH (NOLOCK)
ON clc.wh_id = loc.wh_id
AND clc.location_id = loc.location_id
INNER JOIN t_pick_area pka WITH (NOLOCK)
ON pka.pick_area = loc.pick_area
AND pka.wh_id = loc.wh_id
AND (pka.pick_area <> N'LABEL' OR (pka.pick_area = N'LABEL' AND 0 IS NULL AND 0 IS NULL) )
AND (pka.pick_area_type = 'R' OR (pka.pick_area_type = 'V' and 0 IS NULL) )
INNER JOIN t_zone_loca zlc WITH (NOLOCK)
ON loc.wh_id = zlc.wh_id
AND loc.location_id = zlc.location_id
INNER JOIN (
SELECT loc.wh_id, loc.pnd_location_id --, loc.location_id
FROM t_location loc with (nolock)
inner join t_class_loca clc WITH (NOLOCK)
on clc.location_id = loc.location_id
and clc.wh_id = loc.wh_id
and clc.class_id = 'APPAREL'
LEFT JOIN t_stored_item sto with (nolock)
ON sto.put_away_location = loc.location_id
AND sto.wh_id = loc.wh_id --BTH 20160907 missing wh_id
AND sto.put_away_location IS NOT NULL
AND sto.type = 0
WHERE loc.type in ('I','M')
AND loc.pnd_location_id IS NOT NULL --BTH 20160907 - remove from having clause, add here
GROUP BY loc.wh_id, loc.pnd_location_id, loc.c3
HAVING ((COUNT(sto.hu_id) < 100
and loc.pnd_location_id IS NOT NULL --BTH 201600907
and c3 is null)
OR (COUNT(sto.hu_id) < 500 --and loc.pnd_location_id IS NOT NULL --BTH 201600907
and c3 = 'BULK'))
) as pnd
ON pnd.wh_id = loc.wh_id
AND pnd.pnd_location_id = loc.pnd_location_id
LEFT OUTER JOIN t_put_rules_empty_and_unalloc_locs_by_pnd tpr WITH (NOLOCK)
ON tpr.pnd_location_id = loc.pnd_location_id
AND tpr.class_id = itu.class_id
AND tpr.wh_id = loc.wh_id
LEFT OUTER JOIN t_work_q q WITH(NOLOCK)
ON q.location_id = loc.location_id
AND q.wh_id = loc.wh_id
AND q.work_type = '08'
AND q.work_status = 'U'
WHERE loc.status = 'E'
AND ISNULL(q.work_q_id, 0) = 0
AND ( loc.c3 is null or loc.c3 not in ('R','H','S'))
AND (
(
loc.type = 'M'
AND ( (SELECT TOP 1 max_sku_count
FROM t_zone zone2 (NOLOCK)
WHERE zone2.wh_id = loc.wh_id
AND loc.zone = zone2.zone) >
(SELECT COUNT(sto2.item_number)
FROM t_stored_item sto2 (NOLOCK)
WHERE loc.wh_id = sto2.wh_id
AND loc.location_id = sto2.location_id
)
OR '13098271' IN (SELECT sto2.item_number
FROM t_stored_item sto2 (NOLOCK)
WHERE loc.wh_id = sto2.wh_id
AND loc.location_id = sto2.location_id
)
)
)
OR (loc.type = 'I' AND itu.unit_volume = 0 AND itu.nested_volume = 0)
OR (loc.type = 'I' AND loc.capacity_volume = 0)
OR (loc.type = 'I' AND loc.capacity_volume >= 0 +
(CASE
WHEN 0 = 0 THEN 0
ELSE 0
END * (1 - 1)
)
)
)
--Ensure that only one item is designated to the location
AND (loc.type = 'M'
OR (loc.type = 'I' AND NOT EXISTS (SELECT 1
FROM t_stored_item sto2 WITH (NOLOCK) -- Sum the item in the location to determine volume
WHERE sto2.wh_id = loc.wh_id
AND sto2.put_away_location = loc.location_id)))
AND itu.wh_id = '895'
AND itu.item_number = '13098271'
AND clc.class_id = 'APPAREL'
AND zlc.zone = 'ALL'
ORDER BY
tpr.percent_empty_and_unalloc DESC,
loc.type,
loc.user_count,
loc.picking_flow,
loc.location_id
在不知道您的表和数据的结构的情况下,我的建议是查看执行成本较高的部分。在您的示例中,这将位于右上角:分别为 14%(内部连接)、17% 和 22%,+ 19% 和 25% 其他地方。
其他重要的事情:您的 索引 以及它们是否按应有的方式使用。我觉得不是。
关注 25%(Key Lookup (Clustered)):这 link 可能会帮助您更好地理解问题(并省去冗长的解释)。同样,我不知道你的表的结构。但是我觉得你的索引在这里不够。
我看到了这个:CONVERT_IMPLICIT(nvarchar(4000)
。这是什么 ?会不会降低性能?
如果表的结构不好并且数据模型错误,那么优化起来会更加困难。添加更多索引或改写查询并不总是解决方案。
基于快速浏览您的执行计划。根据成本,第 1 期和第 3 期将为您提供最佳投资回报率。
第一期:Key Lookup
在您的索引 IDX_wh_id_status_pnd_location_id
中,您的索引的 INCLUDE 部分缺少列 zone
。
简而言之:您正在比较不同类型的列。确保比较相同类型的列。如果这些是外键列,则两个表中的类型应该完全相同。如果它们是参数,请更改类型或 cast/covert 它们。
第三期:聚合
您在 [t_stored_item].[sto].put_away_location
上有一个聚合(最大值、计数、平均值...),行估计值为 129108。
尝试创建一个 indexed view for this part of the query with the aggregation in the indexed view. Use the indexed view instead. More info
估计也与实际相差甚远,您可以尝试重建统计数据,但可能无济于事。为什么? Read this
第四期:用户自定义函数
您有一个 INNER JOIN usf_get_item_class_dia_ovrd
是否可以内联编写用户定义函数的逻辑?当我们在内联时,代码通常会被优化,现在标量函数正在逐行执行而不是基于集合。
第五期:常量扫描-实际行数0
这可能不是什么大问题,但当表达式相互抵消时,通常会发生这种情况。虚拟示例:1 = 0
将始终评估为 0 行,因此 SQL 服务器将其替换为空常量扫描。
在复杂的查询中,您可能无法立即找到它。这不会对性能产生很大影响,但是当您从查询中删除这些时,您可能会获得更好的执行计划。
如果有兴趣观看 this video 以更好地了解查询优化器。 (有点旧,但仍然相关)
奖励:参数嗅探
您提到它是一个存储过程。存储过程通常有 parameters sniffing.
的问题