非常慢的查询，包含多个子查询SQL Server 2008 R2

我正在处理一个对表单进行审计的查询。有好几页值得审核的问题。填写表格时，答案以以下方式存储在两个表中：

Table 1: smsmir.obsv OBS
EPISODE NO | FORM USAGE | QUEST      | ANSWER | ...
123456789  | ADMISSION  | QUESTION 1 | YES    | ...
123456789  | ADMISSION  | QUESTION 2 | 150    | ...
...

Table 2: smsdss.QOC_vst_summ QOC
EPISODE NO | HT IND | WT IND | ADV DIR | ...
123456789  |    1   |   1    |     0   | ...
...

表1:

smsmir.obsv OBS

将信息存储在向量中，因此每个问题都有另一行。表2:

smsdss.QOC\u vst\u sum QOC

将答案存储在一行中，因此每次访问只有一行。表3相同，每个访问id只有一行

我的查询首先收集存储在表中的

访问ID

，然后传递到下一组以回答一些问题。我从另一个表中提取访问ID的原因是，访问开始和结束日期存储在该表中。那张桌子看起来像这样：

Table 3: smsdss.BMH_PLM_PtAcct_V PAV
EPISODE NO | ADM DATE   | ...
123456789  | 2013-08-01 | ...
...

我得到的期望输出如下：

EPISODE NO | QUESTION 1  | QUESTION 2 | HT IND | WT IND | ADV DIR | ...
123456789  |   1         | 1          |    1   |    1   |    0    | ...
...

在上表中，1表示使用案例陈述回答了问题，0表示未回答。该查询已被重新编写，现在正在生成正确的结果，但速度非常慢，取回40条记录需要53分36秒。由于查询目前尚未完成，只有7列返回，因此我必须将其扩展为总共65列

我有子查询的原因是答案存储在一个向量中，每一行都是一个问题和答案，但是因为我想在列中显示答案和问题，所以我做了一个子查询。有没有更好的方法来加快速度

以下是查询：

-- THIS QUERY WILL PERFORM AN AUDIT OF THE ADMISSION ASSESSMENT AND
-- OTHER REQUIRED QUESTIONS BY NURSING INFORMATICS
-----------------------------------------------------------------------
-- VARIABLE DECLARATION AND INITIALIZATION. BY DECLARING A START AND
-- END DATE A USER CAN SIMPLY CHANGE THOSE PARAMETERS AND AUDIT ALL 
-- INPATIENT ADMISSION ASSESSMENTS FOR THAT TIME PERIOD
DECLARE @SD DATETIME
DECLARE @ED DATETIME

SET @SD = '2013-08-01'
SET @ED = '2013-08-01'

-- QUERY 1
-- THIS QUERY CREATES A TABLE THAT WILL HOUSE ALL VISIT ID NUMBERS THAT
-- ARE GOING TO BE INCLUDED INSIDE OF THE ADMISSION ASSESSMENT AUDIT
-- TABLE DECLARATION ##################################################
DECLARE @T1 TABLE (
  VISIT_ID VARCHAR(20))

-- ####################################################################
-- THESE ARE THE ITEMS THAT ARE GOING TO BE INSERTED INTO THE TABLE
INSERT INTO @T1
-- COLUMN SELECTION
SELECT A.PtNo_Num
-- DB(S) USED
FROM   (SELECT DISTINCT PTNO_NUM
        FROM   smsdss.BMH_PLM_PtAcct_V
        WHERE  Adm_Date BETWEEN @SD AND @ED
               AND Plm_Pt_Acct_Type = 'I') A

--+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
--+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++//
-----------------------------------------------------------------------
-- QUERY TWO. THIS QUERY WILL TAKE THE VISIT ID'S FROM QUERY 1 AND RUN
-- THEM THROUGH A SET OF RULES TO DECIDE WHEATHER OR NOT THE ADDMISISON
-- ASSESSMENT WAS PROPERLY DONE
-----------------------------------------------------------------------
-- COLUMN SELECTION
SELECT DISTINCT OBS.episode_no  AS [VISIT ID]
                -- CASE STATEMENT, IF PREFERRED LANGUAGE IS NOT 'NULL' THEN CONSIDER
                -- THIS COMPLETE AND SCORE 1 ELSE CONSIDER INCOMPLETE AND SCORE 0
                ,
                CASE
                  WHEN QOC.prim_lng IS NOT NULL THEN 1
                  ELSE 0
                END             AS [PREF LANG COMPLETE?],
                QOC.ht_chtd_ind AS [HT IND],
                QOC.wt_chtd_ind AS [WT IND],
                QOC.adv_dir_ind AS [ADV DIRECTIVE]
                -- A SEPERATE SELECT STATEMENT IS USED HERE BECAUSE RESULTS OF THE
                -- ADMISSION CONSENT ARE STORED IN A VECTOR, SO IT IS NECESSARY TO
                -- MAKE A SELECTION FROM THAT LIST, HERE A VALUE OF 1 = YES AND 
                -- 0 = NO
                ,
                CASE
                  WHEN OBS.episode_no NOT IN (SELECT episode_no
                                              FROM   smsmir.obsv
                                              WHERE  form_usage = 'Admission') THEN 0
                  ELSE 1
                END             AS [ADMIT ASSESSMENT DONE],
                CASE
                  WHEN OBS.episode_no NOT IN (SELECT episode_no
                                              FROM   smsmir.obsv
                                              WHERE  form_usage = 'Admission'
                                                     AND obsv_cd_ext_name = 'Admission consent signed:') THEN 0
                  ELSE 1
                END             AS [ADMIT CONSENT SIGNED?]
-- DB(S) USED ---------------------------------------------------------
FROM   smsmir.obsv OBS
       JOIN smsdss.QOC_vst_summ QOC
         ON OBS.episode_no = QOC.episode_no
       JOIN @T1 T1
         ON OBS.episode_no = T1.VISIT_ID
-- FILTERS ------------------------------------------------------------
WHERE  T1.VISIT_ID = OBS.episode_no
GROUP  BY OBS.episode_no,
          QOC.prim_lng,
          QOC.ht_chtd_ind,
          QOC.wt_chtd_ind,
          QOC.adv_dir_ind,
          OBS.obsv_cd_ext_name
--#####################################################################
-- END REPORT ...[]...[]...[]

您会注意到，我使用的是

不在

子句中，原因是如果一个问题没有被问到或回答，就不会有记录，甚至不会有一个

NULL

，所以如果我不使用它，这个人可以完成所有其他事情，但如果不是那个特定的项目，那么他们就被排除在最终结果集中

如果我需要澄清，请让我知道

**查询实际执行计划XML**

---

谢谢

smsmir。obsv

是一个视图，它包含一个155569000行的表和一个15375000行的表

执行计划显示这些表被扫描了42次

这绝大多数是因为表变量的默认基数估计值很低，这意味着嵌套循环的选择不恰当。用

#temp

表替换应该可以解决这个问题

另外，使用

PIVOT

技术而不是单独的子查询可以进一步减少这种情况。在添加缺少的索引方面，可能会有额外的优化，但您能否尝试一下，并让我知道时间安排和执行计划

DECLARE @SD DATETIME = '2013-08-01';
DECLARE @ED DATETIME = '2013-08-01';

CREATE TABLE #T1
  (
     VISIT_ID VARCHAR(20) UNIQUE CLUSTERED
  )

INSERT INTO #T1
SELECT DISTINCT PTNO_NUM
FROM   smsdss.BMH_PLM_PtAcct_V
WHERE  Adm_Date BETWEEN @SD AND @ED
       AND Plm_Pt_Acct_Type = 'I'
OPTION (RECOMPILE);

WITH OBS
     AS (SELECT episode_no,
                MAX(CASE
                      WHEN form_usage = 'Admission' THEN 1
                    END) AS [ADMIT ASSESSMENT DONE],
                MAX(CASE
                      WHEN form_usage = 'Admission'
                           AND obsv_cd_ext_name = 'Admission consent signed:' THEN 1
                    END) AS [ADMIT CONSENT SIGNED?]
         FROM   smsmir.obsv
         WHERE form_usage = 'Admission' 
         GROUP  BY episode_no)
SELECT OBS.episode_no                         AS [VISIT ID],
       CASE
         WHEN QOC.prim_lng IS NOT NULL THEN 1
         ELSE 0
       END                                    AS [PREF LANG COMPLETE?],
       QOC.ht_chtd_ind                        AS [HT IND],
       QOC.wt_chtd_ind                        AS [WT IND],
       QOC.adv_dir_ind                        AS [ADV DIRECTIVE],
       ISNULL(OBS.[ADMIT ASSESSMENT DONE], 0) AS [ADMIT ASSESSMENT DONE],
       ISNULL(OBS.[ADMIT CONSENT SIGNED?], 0) AS [ADMIT CONSENT SIGNED?]
FROM   smsdss.QOC_vst_summ QOC
       JOIN #T1
         ON #T1.VISIT_ID = QOC.episode_no
       LEFT JOIN OBS
         ON OBS.episode_no = QOC.episode_no 

DROP TABLE #T1

---

smsmir.obsv

是一个视图，它包含一个155569000行的表和一个15375000行的表

执行计划显示这些表被扫描了42次

这绝大多数是因为表变量的默认基数估计值很低，这意味着嵌套循环的选择不恰当。用

#temp

表替换应该可以解决这个问题

另外，使用

PIVOT

技术而不是单独的子查询可以进一步减少这种情况。在添加缺少的索引方面，可能会有额外的优化，但您能否尝试一下，并让我知道时间安排和执行计划

DECLARE @SD DATETIME = '2013-08-01';
DECLARE @ED DATETIME = '2013-08-01';

CREATE TABLE #T1
  (
     VISIT_ID VARCHAR(20) UNIQUE CLUSTERED
  )

INSERT INTO #T1
SELECT DISTINCT PTNO_NUM
FROM   smsdss.BMH_PLM_PtAcct_V
WHERE  Adm_Date BETWEEN @SD AND @ED
       AND Plm_Pt_Acct_Type = 'I'
OPTION (RECOMPILE);

WITH OBS
     AS (SELECT episode_no,
                MAX(CASE
                      WHEN form_usage = 'Admission' THEN 1
                    END) AS [ADMIT ASSESSMENT DONE],
                MAX(CASE
                      WHEN form_usage = 'Admission'
                           AND obsv_cd_ext_name = 'Admission consent signed:' THEN 1
                    END) AS [ADMIT CONSENT SIGNED?]
         FROM   smsmir.obsv
         WHERE form_usage = 'Admission' 
         GROUP  BY episode_no)
SELECT OBS.episode_no                         AS [VISIT ID],
       CASE
         WHEN QOC.prim_lng IS NOT NULL THEN 1
         ELSE 0
       END                                    AS [PREF LANG COMPLETE?],
       QOC.ht_chtd_ind                        AS [HT IND],
       QOC.wt_chtd_ind                        AS [WT IND],
       QOC.adv_dir_ind                        AS [ADV DIRECTIVE],
       ISNULL(OBS.[ADMIT ASSESSMENT DONE], 0) AS [ADMIT ASSESSMENT DONE],
       ISNULL(OBS.[ADMIT CONSENT SIGNED?], 0) AS [ADMIT CONSENT SIGNED?]
FROM   smsdss.QOC_vst_summ QOC
       JOIN #T1
         ON #T1.VISIT_ID = QOC.episode_no
       LEFT JOIN OBS
         ON OBS.episode_no = QOC.episode_no 

DROP TABLE #T1

---

你不必再敲桌子了，你已经有数据了。把它加起来

DECLARE @SD DATETIME
DECLARE @ED DATETIME

SET @SD = '2013-08-01'
SET @ED = '2013-08-01'

DECLARE @T1 TABLE (
  VISIT_ID VARCHAR(20))

INSERT INTO @T1
SELECT A.PtNo_Num
FROM   (SELECT DISTINCT PTNO_NUM
        FROM   smsdss.BMH_PLM_PtAcct_V
        WHERE  Adm_Date BETWEEN @SD AND @ED
           AND Plm_Pt_Acct_Type = 'I') A

SELECT DISTINCT OBS.episode_no  AS [VISIT ID],
            CASE
              WHEN QOC.prim_lng IS NOT NULL THEN 1
              ELSE 0
            END             AS [PREF LANG COMPLETE?],
            QOC.ht_chtd_ind AS [HT IND],
            QOC.wt_chtd_ind AS [WT IND],
            QOC.adv_dir_ind AS [ADV DIRECTIVE],
            max(CASE
              WHEN   form_usage = 'Admission' THEN 1
              ELSE 0
            END)             AS [ADMIT ASSESSMENT DONE],
            max(CASE
              WHEN   form_usage = 'Admission' AND obsv_cd_ext_name = 'Admission consent signed:' THEN 1
              ELSE 0
            END )            AS [ADMIT CONSENT SIGNED?]
FROM   smsmir.obsv OBS
   JOIN smsdss.QOC_vst_summ QOC
     ON OBS.episode_no = QOC.episode_no
   JOIN @T1 T1
     ON OBS.episode_no = T1.VISIT_ID
WHERE  T1.VISIT_ID = OBS.episode_no
GROUP  BY OBS.episode_no,
      QOC.prim_lng,
      QOC.ht_chtd_ind,
      QOC.wt_chtd_ind,
      QOC.adv_dir_ind,
      OBS.obsv_cd_ext_name

---

你不必再敲桌子了，你已经有数据了。把它加起来

DECLARE @SD DATETIME
DECLARE @ED DATETIME

SET @SD = '2013-08-01'
SET @ED = '2013-08-01'

DECLARE @T1 TABLE (
  VISIT_ID VARCHAR(20))

INSERT INTO @T1
SELECT A.PtNo_Num
FROM   (SELECT DISTINCT PTNO_NUM
        FROM   smsdss.BMH_PLM_PtAcct_V
        WHERE  Adm_Date BETWEEN @SD AND @ED
           AND Plm_Pt_Acct_Type = 'I') A

SELECT DISTINCT OBS.episode_no  AS [VISIT ID],
            CASE
              WHEN QOC.prim_lng IS NOT NULL THEN 1
              ELSE 0
            END             AS [PREF LANG COMPLETE?],
            QOC.ht_chtd_ind AS [HT IND],
            QOC.wt_chtd_ind AS [WT IND],
            QOC.adv_dir_ind AS [ADV DIRECTIVE],
            max(CASE
              WHEN   form_usage = 'Admission' THEN 1
              ELSE 0
            END)             AS [ADMIT ASSESSMENT DONE],
            max(CASE
              WHEN   form_usage = 'Admission' AND obsv_cd_ext_name = 'Admission consent signed:' THEN 1
              ELSE 0
            END )            AS [ADMIT CONSENT SIGNED?]
FROM   smsmir.obsv OBS
   JOIN smsdss.QOC_vst_summ QOC
     ON OBS.episode_no = QOC.episode_no
   JOIN @T1 T1
     ON OBS.episode_no = T1.VISIT_ID
WHERE  T1.VISIT_ID = OBS.episode_no
GROUP  BY OBS.episode_no,
      QOC.prim_lng,
      QOC.ht_chtd_ind,
      QOC.wt_chtd_ind,
      QOC.adv_dir_ind,
      OBS.obsv_cd_ext_name

---

请将实际（非估计）执行计划的XML上传到某个地方。如果可以的话，运行它大约需要30分钟左右的时间，以便显示需要子查询的两个问题。在这种情况下，请尝试用

创建表#T1替换表变量
@T1
（访问ID VARCHAR（20）主键）

。这可能会有帮助，但至少不应该妨碍。@MartinSmith现在在一个单独的查询窗口中尝试，您还提到答案存储在两个表中，但随后向我们显示了三个，表和列的名称与查询中使用的名称不符。目前还不清楚预期结果如何与示例数据相对应。如果您提供更广泛的示例数据和结果，您可能会得到更好的答案。请将实际（非估计）执行计划的XML上载到某处。可以，为了显示需要子查询的两个问题，运行它大约需要30分钟。在这种情况下，请尝试将表变量

@T1

替换为

CREATE table#T1（VISIT_ID VARCHAR（20）主键）

。这可能会有帮助，但至少不应该妨碍。@MartinSmith现在在一个单独的查询窗口中尝试，您还提到答案存储在两个表中，但随后向我们显示了三个，表和列的名称与查询中使用的名称不符。目前还不清楚预期结果如何与示例数据相对应。如果你提供更广泛的示例数据和结果，你可能会得到一个更好的答案。我现在就试试，因为某种原因，我昨晚放学回家时从未被通知你发布了答案，现在就开始运行并发布结果。好吧，这非常快（相对于我所拥有的，0:4:56:00），但是当一条记录没有结果时，我会得到一个访问ID的空值，而不是像123456789这样的实际ID。获取ID号是非常重要的，尤其是在存在0的情况下，因为我们想调查为什么没有执行这些操作。如果我们能解决这个问题，那么这个问题就会被标记为“已回答”，你就会得到分数。这是新的实际执行XML@MCP_infirator-哦，是的。您需要的不是

SELECT OBS.eposion\u no

而是

SELECT\T1.VISIT\u ID

@MCP\u infirator-您的完整查询是否只对表格用法为“准入”的行

感兴趣