查询中的几个连接 - 可能替换以获得性能？

Question

我有一个包含1000万行的表,我试图找到谁是某些机器(id)的第一个/最后一个维护者,具体取决于某些日期,也取决于机器的状态.我的查询使用了六个连接,还有其他首选选项吗？ 编辑:原始表有索引,试图优化查询替换连接 - 如果可能的话？SQL小提琴示例:

SQL小提琴

编辑(在下面添加了其他信息):

示例表:

CREATE TABLE vendor_info (
  id INT,
  datestamp INT,
  statuz INT,
  maintainer VARCHAR(25));

  INSERT INTO vendor_info VALUES (1, 20180101, 0, 'Jay');
  INSERT INTO vendor_info VALUES (2, 20180101, 0, 'Eric');
  INSERT INTO vendor_info VALUES (3, 20180101, 1, 'David');
  INSERT INTO vendor_info VALUES (1, 20180201, 1, 'Jay');
  INSERT INTO vendor_info VALUES (2, 20180201, 0, 'Jay');
  INSERT INTO vendor_info VALUES (3, 20180201, 1, 'Jay');
  INSERT INTO vendor_info VALUES (1, 20180301, 1, 'Jay');
  INSERT INTO vendor_info VALUES (2, 20180301, 1, 'David');
  INSERT INTO vendor_info VALUES (3, 20180301, 1, 'Eric');

查询和所需的输出:

SELECT
   id
  , MIN(datestamp) AS min_datestamp
  , MAX(datestamp) AS max_datestamp
  , MAX(case when statuz = 0 then datestamp end) AS max_s0_date
  , MAX(case when statuz = 1 then datestamp end) AS max_s1_date
  , MIN(case when statuz = 0 then datestamp end) AS min_s0_date
  , MIN(case when statuz = 1 then datestamp end) AS min_s1_date
 INTO vendor_dates
 FROM vendor_info
 GROUP BY id;

SELECT 
    vd.id
  , v1.maintainer  AS first_maintainer
  , v2.maintainer AS last_maintainer 
  , v3.maintainer AS last_s0_maintainer
  , v4.maintainer AS last_s1_maintainer
  , v5.maintainer AS first_s0_maintainer
  , v6.maintainer AS first_s1_maintainer

FROM vendor_dates vd
LEFT JOIN vendor_info v1 ON vd.id = v1.id AND vd.min_datestamp = v1.datestamp
LEFT JOIN vendor_info v2 ON vd.id = v2.id AND vd.max_datestamp = v2.datestamp
LEFT JOIN vendor_info v3 ON vd.id = v3.id AND vd.max_s0_date = v3.datestamp
LEFT JOIN vendor_info v4 ON vd.id = v4.id AND vd.max_s1_date = v4.datestamp
LEFT JOIN vendor_info v5 ON vd.id = v5.id AND vd.min_s0_date = v5.datestamp
LEFT JOIN vendor_info v6 ON vd.id = v6.id AND vd.min_s1_date = v6.datestamp;

Answer 1

Adding an index to vendor_info reduces duration of your 2nd query from over 300ms to under 30ms average over repeated runs

PRIMARY KEY CLUSTERED (id, datestamp)

Changing the 2 step process into a CTE reduces total duration even more to well under 15ms average over repeated runs.

The CTE method lets the query optimiser use the new primary key

CREATE TABLE vendor_info (
  id INT,
  datestamp INT,
  statuz INT,
  maintainer VARCHAR(25)

  PRIMARY KEY CLUSTERED (id, datestamp)
);

  INSERT INTO vendor_info VALUES (1, 20180101, 0, 'Jay');
  INSERT INTO vendor_info VALUES (2, 20180101, 0, 'Eric');
  INSERT INTO vendor_info VALUES (3, 20180101, 1, 'David');
  INSERT INTO vendor_info VALUES (1, 20180201, 1, 'Jay');
  INSERT INTO vendor_info VALUES (2, 20180201, 0, 'Jay');
  INSERT INTO vendor_info VALUES (3, 20180201, 1, 'Jay');
  INSERT INTO vendor_info VALUES (1, 20180301, 1, 'Jay');
  INSERT INTO vendor_info VALUES (2, 20180301, 1, 'David');
  INSERT INTO vendor_info VALUES (3, 20180301, 1, 'Eric');

WITH vendor_dates AS
(SELECT
   id
  , MIN(datestamp) AS min_datestamp
  , MAX(datestamp) AS max_datestamp
  , MAX(case when statuz = 0 then datestamp end) AS max_s0_date
  , MAX(case when statuz = 1 then datestamp end) AS max_s1_date
  , MIN(case when statuz = 0 then datestamp end) AS min_s0_date
  , MIN(case when statuz = 1 then datestamp end) AS min_s1_date
 FROM vendor_info
 GROUP BY id
 )
 SELECT 
    vd.id
  , v1.maintainer  AS first_maintainer
  , v2.maintainer AS last_maintainer 
  , v3.maintainer AS last_s0_maintainer
  , v4.maintainer AS last_s1_maintainer
  , v5.maintainer AS first_s0_maintainer
  , v6.maintainer AS first_s1_maintainer

FROM vendor_dates vd
LEFT JOIN vendor_info v1 ON vd.id = v1.id AND vd.min_datestamp = v1.datestamp
LEFT JOIN vendor_info v2 ON vd.id = v2.id AND vd.max_datestamp = v2.datestamp
LEFT JOIN vendor_info v3 ON vd.id = v3.id AND vd.max_s0_date = v3.datestamp
LEFT JOIN vendor_info v4 ON vd.id = v4.id AND vd.max_s1_date = v4.datestamp
LEFT JOIN vendor_info v5 ON vd.id = v5.id AND vd.min_s0_date = v5.datestamp
LEFT JOIN vendor_info v6 ON vd.id = v6.id AND vd.min_s1_date = v6.datestamp;