如何提高下面提到的 SQL 查询中的文本列性能
Posted
技术标签:
【中文标题】如何提高下面提到的 SQL 查询中的文本列性能【英文标题】：How to improve the text column performance in below mentioned SQL query 【发布时间】：2017-08-14 10:58:40 【问题描述】：有一个 SQL union all 查询和 3 个 union all 查询。在查询中添加文本列CAST(c.getQuestionId AS VARCHAR(300)) 后，查询执行时间发生了显着变化。使用的数据库是SQL SERVER 2014，请性能调优高手帮忙。
JcccustomersAssessmentProxy 上创建了 3 个索引。这是一个包含 8000 万条记录的表。仅在第三个查询中提到显式使用索引
（即（NOLOCK, INDEX=IX_AssessmentProxy_myJcAssessmentContext））
myJcAssessmentContext 上有一个非聚集索引
getmycoordtoyplanver 上有一个非聚集索引
getMyEventItem 上有一个非聚集索引
SELECT ass.P_KEY
,ass.SS_CODE
,CAST(evp.EVENT_SSID AS VARCHAR(11)) AS EVENT_SSID
,CAST(QUESTIONNUMBER AS VARCHAR(3)) AS EVENT_NO
,CAST(ISNULL(CAST(TEMPLATEVERSION AS VARCHAR), 0) AS INT) AS TEMPLATEVERSION
,ISNULL(CAST(TEMPLATENAME AS VARCHAR(50)), ' ') AS TEMPLATENAME
,evp.ASS_DATE
,ISNULL(ANSWERNUMBER, 0) AS ANSWER_NO
,ass.CASE_SSID
,ass.RE_SSID
,ass.RE_DATE
,ass.EPISODE_SSID
,ass.[SERVICE]
,ass.SERVICE_DESC
,ass.TAM_KEY
,ass.PRv_KEY
,CAST(QUESTIONNUMBER AS VARCHAR(3)) AS QUESTION_NO
,ISNULL(CONVERT(VARCHAR(100), REPLACE(REPLACE(CAST(QUESTIONTEXT AS VARCHAR(650)), char(10), ''), char(13), ' ')), ' ') AS QUESTION_TEXT
,ISNULL(CAST(ANSWER AS VARCHAR(125)), ' ') AS ANSWER_TEXT
,a.MyAssessment AS ASSESSMENT_SSID
-----------------------------------------------------------------------------------------
,CAST(RIGHT(a.getMyCoordtoyPlanVer, 10) AS INT) AS toy_PLAN_VERSION_SSID
,CAST(RIGHT(a.myJcAssessmentContext, 10) AS INT) AS CONTEXT_FORM_SSID
-----------------------------------------------------------------------------------------
,ass.RECORD_DATE
**,CAST(a.getQuestionId AS VARCHAR(300) ) AS QUESTION_ID /*NEW Text Column ,this remove the parallelism*/**
FROM SourceFeed.dbo.JcccustomersAssessmentProxy AS a WITH(NOLOCK)
INNER JOIN
TEMP_DATABASE.dbo.jsystemReport_AllEventItems_AllPersons evp WITH(NOLOCK)
ON a.getMyEventItem = evp.oid
INNER JOIN
TEMP_DATABASE.dbo.jsystemAssessment ass WITH(NOLOCK)
ON evp.getcustomersId = ass.customersID
AND evp.EVENT_SSID = ass.EVENTID
UNION ALL
SELECT ass.P_KEY
,ass.SS_CODE
,CAST(csp.ENT_SSID AS VARCHAR(11)) AS EVENT_SSID
,CAST(jcc.QUESTIONNUMBER AS VARCHAR(3)) AS EVENT_NO
,CAST(ISNULL(CAST(CAST(jcc.TEMPLATEVERSION AS VARCHAR) AS VARCHAR), 0) AS INT) AS TEMPLATEVERSION
,ISNULL(CAST(jcc.TEMPLATENAME AS VARCHAR(50)), ' ') AS TEMPLATENAME
,csp.ASSESSMENT_DATE
,ISNULL(jcc.ANSWERNUMBER, 0) AS ANSWER_NO
,ass.CASE_SSID
,ass.REF_SSID
,ass.RE_DATE
,ass.EPISODE_SSID
,ass.[SERVICE]
,ass.SERVICE_DESC
,ass.TEAM_KEY
,ass.PROVIDER_KEY
,CAST(ISNULL(jcc.QUESTIONNUMBER, 0) AS VARCHAR(3)) AS QUESTION_NO
,isnull(CONVERT(VARCHAR(100), replace(replace(CAST(jcc.QUESTIONTEXT AS VARCHAR(650)), char(10), ''), char(13), ' ')), ' ') AS QUESTION_TEXT
,ISNULL(CAST(jcc.ANSWER AS VARCHAR(125)), ' ') AS ANSWER_TEXT
,jcc.MyAssessment AS ASSESSMENT_SSID
-----------------------------------------------------------------------------------------
,CAST(RIGHT(jcc.getMyCoordtoyPlanVer, 10) AS INT) AS toy_PLAN_VERSION_SSID
,CAST(RIGHT(jcc.myJcAssessmentContext, 10) AS INT) AS CONTEXT_FORM_SSID
-----------------------------------------------------------------------------------------
,ass.RECORD_DATE
**,CAST(jcc.getQuestionId AS VARCHAR(300) ) AS QUESTION_ID /*NEW Text Column ,this remove the parallelism*/**
FROM TEMP_DATABASE.dbo.jsystemReport_toySpell csp WITH(NOLOCK)
INNER JOIN
TEMP_DATABASE.dbo.jsystemAssessment ass WITH(NOLOCK)
ON csp.getcustomersId = ass.customersID
AND csp.EVENT_SSID = ass.EVENTID
LEFT OUTER JOIN
SourceFeed.dbo.JcccustomersAssessmentProxy jcc WITH(NOLOCK)
ON csp.OID = jcc.getmycoordtoyplanver
WHERE NOT EXISTS (SELECT tp1.OID
FROM TEMP_DATABASE.dbo.AssessmentTransferPart1 tp1
WHERE tp1.OID = jcc.OID)
UNION ALL
SELECT ass.P_KEY
,ass.SS_CODE
,CAST(a.EVENT_SSID AS VARCHAR(11)) AS EVENT_SSID
,CAST(CAST(C.QUESTIONNUMBER AS INT) AS CHAR(3)) AS EVENT_NO
,CAST(ISNULL(CAST(C.TEMPLATEVERSION AS VARCHAR), 0) AS INT) AS TEMPLATEVERSION
,ISNULL(CAST(C.TEMPLATENAME AS VARCHAR(50)), ' ') AS TEMPLATENAME
,a.ASSESSMENT_DATE
,ISNULL(C.ANSWERNUMBER, 0) AS ANSWER_NO
,ass.CASE_SSID
,ass.REL_SSID
,ass.REAL_DATE
,ass.EPISODE_SSID
,ass.[SERVICE]
,ass.SERVICE_DESC
,ass.TEAM_KEY
,ass.PROVIDER_KEY
,CAST(ISNULL(CAST(C.QUESTIONNUMBER AS INT), 0) AS VARCHAR(3)) AS QUESTION_NO
,isnull(CONVERT(VARCHAR(100), CAST(C.QUESTIONTEXT AS VARCHAR(650))), ' ') AS QUESTION_TEXT
,ISNULL(CAST(C.ANSWER AS VARCHAR(125)), ' ') AS ANSWER_TEXT
,a.MyAssessment AS ASSESSMENT_SSID
-----------------------------------------------------------------------------------------
,CAST(RIGHT(C.getMyCoordtoyPlanVer, 10) AS INT) AS toy_PLAN_VERSION_SSID
,CAST(RIGHT(C.myJcAssessmentContext, 10) AS INT) AS CONTEXT_FORM_SSID
-----------------------------------------------------------------------------------------
,ass.RECORD_DATE
**,CAST(c.getQuestionId AS VARCHAR(300) )** AS QUESTION_ID /*NEW Text Column ,this remove the parallelism*/
FROM SourceFeed.dbo.toyClusterReviewAssessment B WITH(NOLOCK)
INNER JOIN
SourceFeed.dbo.JcccustomersAssessmentProxy C WITH(NOLOCK, INDEX=IX_AssessmentProxy_myJcAssessmentContext)
ON B.myJCJccUserFormContext = C.myJcAssessmentContext
INNER JOIN
SourceFeed.dbo.JcccustomersAssessmentScoreProxy D WITH(NOLOCK)
ON B.myJCJccUserFormContext = D.myJcAssessmentContext
INNER JOIN
TEMP_DATABASE.dbo.jsystemReport_toyClusterReviewEvent_AllPersons A WITH(NOLOCK)
ON a.myAssessment = B.oid
INNER JOIN
TEMP_DATABASE.dbo.jsystemAssessment ass WITH(NOLOCK)
ON a.getcustomersId = ass.customersID
AND a.EVENT_SSID = ass.EVENTID
WHERE NOT EXISTS (SELECT OID
FROM (SELECT tp1.OID
FROM TEMP_DATABASE.dbo.AssessmentTransferPart1 tp1
UNION ALL
SELECT tp2.OID
FROM TEMP_DATABASE.dbo.AssessmentTransferPart2 tp2) jcc
WHERE jcc.OID = C.OID)
这里是link to the actual execution plan。
在 JccClientAssessmentProxy jcc 添加索引之后 WITH(NOLOCK,INDEX=IX_AssessmentProxy_GETMYCOORDCAREPLANVER)
这里是the new execution plan。
【问题讨论】：
您添加了哪个文本列？您刚刚在 SELECT 子句中添加了一列，它会减慢您的查询速度？你必须更具体，这样你的问题不是很容易理解。可能添加带有表结构等的 SQL Fiddle。 getQuestionId 是列 注意粗体等标记在格式化的代码块中不起作用。 你应该在这里发布一个执行计划，有a website that allows for posting complicated plans不需要发布图像。 您在JccClientAssesmentProxy 上进行了多次表扫描和一次 RID 查找，这需要 30.8% 的时间。您应该改进该表的索引。 SSMS Execution plan 给出了一些索引建议，请仔细查看并修改您的索引。
【参考方案1】：
有问题的部分是：
,CAST(a.getQuestionId AS VARCHAR(300)) AS QUESTION_ID
FROM cnlPjccR_Report.dbo.JccClientAssessmentProxy AS a
WITH(NOLOCK,INDEX=IX_AssessmentProxy_getMyEventItem)
INNER JOIN ...
并且问题在您的两个 UNION ALL 子查询中重复出现。
您的执行计划显示成本很高Key lookups。
第二种方法是看看你是否可以创建一个“覆盖索引” 满足整个查询或至少消除关键查找。 A “覆盖索引”只是一个非聚集索引，它包含所有 满足整个查询或在我们的例子中需要的列， 消除对密钥查找操作的需要。挑战之一是获得 生成键查找的列列表。你可以这样做 在 SQL Server Management Studio (SSMS) 中通过右键单击键 查找运算符，然后选择属性。然后找到输出 在“属性”窗口中列出行，然后单击省略号按钮。 这将打开一个窗口（见下文），其中包含所有列的列表 键查找正在寻找。您可以使用此列表来帮助您 决定是否以及如何创建索引来“覆盖”查询或键 查找。
在您的情况下，应包含以下列以创建覆盖索引：
[cnlPjccR_Report].[dbo].[JccClientAssessmentProxy].answer;
[cnlPjccR_Report].[dbo].[JccClientAssessmentProxy].answerNumber;
[cnlPjccR_Report].[dbo].[JccClientAssessmentProxy].getMyCoordCarePlanVer;
[cnlPjccR_Report].[dbo].[JccClientAssessmentProxy].getQuestionId;
[cnlPjccR_Report].[dbo].[JccClientAssessmentProxy].myAssessment;
[cnlPjccR_Report].[dbo].[JccClientAssessmentProxy].myJcAssessmentContext;
[cnlPjccR_Report].[dbo].[JccClientAssessmentProxy].questionNumber;
[cnlPjccR_Report].[dbo].[JccClientAssessmentProxy].questionText;
[cnlPjccR_Report].[dbo].[JccClientAssessmentProxy].templateName;
[cnlPjccR_Report].[dbo].[JccClientAssessmentProxy].templateVersion
您的索引可能不包含 getQuestionId 列，而这一事实在您的查询执行计划中造成了较慢的 Key lookup 或 RID lookup。
见also here 或this answer。
【讨论】：
以上是关于如何提高下面提到的 SQL 查询中的文本列性能的主要内容，如果未能解决你的问题，请参考以下文章