Oracle SQL - 多个 JOINS 性能问题

Question

我目前正在尝试为 Oracle DB 编写查询，以获取有关患者住院时间的一些信息。对于他每天在特定领域获得特定分数 - 因此我必须加入多个表格才能获得实际的每日分数（而不是完整分数）。

我的第一次尝试（这是有效的，但我想以另一种格式显示）是这样的：

SELECT
    f.FALLID AS "ID",
    f.FALLNR AS "Fallnummer",
    DECODE(e.SCOREID, 12, 'SAPS', 13, 'TISS', 'X') AS "Score",
    to_char(e.VON, 'DD.MM.YYYY') AS "Datum (von)",
    -- to_char(e.BIS, 'DD.MM.YYYY') AS "Datum (bis)",
    scp.BEZEICHNUNG AS "Bezeichnung",
    s.PUNKTE AS "Punkte"
FROM
    FALL f
    JOIN OS_MUP.EINSTUFUNG e ON e.FALLID = f.FALLID
    JOIN OS_MUP.EINSTUFUNGSRANG einr ON einr.EINSTUFUNGID = e.EINSTUFUNGID
    JOIN OS_MUP.SCORERANG s ON s.SCORERANGID = einr.SCORERANGID
    JOIN OS_MUP.SCOREPOSITION scp ON scp.SCOREPOSITIONID = s.SCOREPOSITIONID
WHERE
    f.FALLNR = 1234567
ORDER BY
    e.SCOREID, e.VON

这以以下形式显示结果：

+------+---------+-------+------------+-------------+--------+
|  ID  | Fallnr  | Score |    Date    | Description | Points |
+------+---------+-------+------------+-------------+--------+
| 1234 | 1234567 | ABCD  | 05.05.2020 | Age         |     12 |
| 1234 | 1234567 | ABCD  | 05.05.2020 | Temperature |      2 |
+------+---------+-------+------------+-------------+--------+

没关系，但我不想重复列的日期和每天的描述，我想得到这样的外观：

+------+---------+-------+------------+-----+-------------+
|  ID  | Fallnr  | Score |    Date    | Age | Temperature |
+------+---------+-------+------------+-----+-------------+
| 1234 | 1234567 | ABCD  | 05.05.2020 |  12 |           2 |
| 1234 | 1234567 | ABCD  | 06.05.2020 |  12 |           2 |
+------+---------+-------+------------+-----+-------------+

所以我修改了查询并在此刻完成了这个：

variable var number
exec :var := 2068237610

SELECT
    dt1.FALLNR AS "Fallnummer",
    to_char(dt1.VON, 'DD.MM.YYYY') AS "Datum",
    dt1.PUNKTE AS "Alter",
    dt2.PUNKTE AS "Herzfrequenz"
    --dt3.PUNKTE AS "RR Syst.",
    --dt4.PUNKTE AS "Körpertemp.",
    --dt5.PUNKTE AS "PaO2/FiO2"
    -- dt6.PUNKTE AS "Urin",
    -- dt7.PUNKTE AS "Leukozyten"
    FROM (
    SELECT
        f.FALLID,
        f.FALLNR,
        s.PUNKTE,
        e.VON
    FROM Fall f
        JOIN OS_MUP.EINSTUFUNG e ON e.FALLID = f.FALLID
        JOIN OS_MUP.EINSTUFUNGSRANG einr ON einr.EINSTUFUNGID = e.EINSTUFUNGID
        JOIN OS_MUP.SCORERANG s ON s.SCORERANGID = einr.SCORERANGID
    WHERE
        s.SCOREPOSITIONID = 10003 AND f.FALLNR = :var /* Alter */
    ) dt1
JOIN
    (SELECT
        f.FALLNR,
        s.PUNKTE,
        e.VON
    FROM Fall f
        JOIN OS_MUP.EINSTUFUNG e ON e.FALLID = f.FALLID
        JOIN OS_MUP.EINSTUFUNGSRANG einr ON einr.EINSTUFUNGID = e.EINSTUFUNGID
        JOIN OS_MUP.SCORERANG s ON s.SCORERANGID = einr.SCORERANGID
    WHERE
        s.SCOREPOSITIONID = 10004 AND f.FALLNR = :var /* Herzfrequenz */
    ) dt2
ON dt1.FALLNR = dt2.FALLNR AND dt1.VON = dt2.VON
ORDER BY dt1.VON

这是我希望它工作的工作，但我必须一遍又一遍地进行这些连接以获得更多列，并且还有 5-6 更多列 - 以及由创建的嵌套循环的原因加入这个版本的查询非常慢（比如慢了 250 秒）。

是否有机会以更少使用联接的另一种方式实现我的目标？

提前致谢！

Answer 1

如何进行转置

一个选项是枢轴：https://www.oracletutorial.com/oracle-basics/oracle-pivot/ 或更多选项：https://livesql.oracle.com/apex/livesql/file/tutorial_GNZ3LQPJ0K6RTD1NEEPNRQT0R.html

主要缺点：

• 直接 SQL 的固定项目列表
• 仅动态项目的 XML 输出

然后是老式的方式：

• 先准备好数据
• 折叠列以输出和分配
• 群

类似的东西：

WITH
  cte_position
(
  key_position,
  name_position
)
AS
( -- manual preparation of column name list
  SELECT Cast( 10003 AS number(6,0) ) AS key_position, Cast( 'Alter' AS varchar2(20 char) ) AS name_position FROM dual
  UNION ALL
  SELECT Cast( 10004 AS number(6,0) ) AS key_position, Cast( 'Herzfrequenz' AS varchar2(20 char) ) AS name_position FROM dual
),
  cte_data
(
  fallnr,
  punkte,
  column_group,
  von
)
AS
( -- combine data and prepare for single pass data collection
  SELECT
    f.FALLNR,
    e.VON,
    punkt_name.name_position AS column_group,
    s.PUNKTE
  FROM Fall f
    INNER JOIN OS_MUP.EINSTUFUNG e 
      ON  e.FALLID = f.FALLID
    INNER JOIN OS_MUP.EINSTUFUNGSRANG einr 
      ON  einr.EINSTUFUNGID = e.EINSTUFUNGID
    INNER JOIN OS_MUP.SCORERANG s 
      ON  s.SCORERANGID = einr.SCORERANGID
    -- link (and filter) by required positions to show
    INNER JOIN cte_position punkt_name
      ON  s.SCOREPOSITIONID = punkt_name.key_position
  WHERE -- specific patient
      f.FALLNR = :var 
)
SELECT
  patient.fallnr AS Fallnummer,
  To_char( patient.von, 'DD.MM.YYYY' ) AS Datum, -- non ISO 8601 date
  Max( CASE WHEN patient.column_group = 'Alter' THEN patient.punkte ELSE 0 END ) AS "Punkt Alter",
  Max( CASE WHEN patient.column_group = 'Herzfrequenz' THEN patient.punkte ELSE 0 END ) AS "Punkt Herzfrequenz"
FROM
  cte_data patient
GROUP BY
  patient.fallnr,
  patient.von
;

同样，这种方法也有一些缺点：

• 每个 Max 函数都必须单独命名

另一方面，这是一次捕获数据的过程，理论上可以加快数据收集速度。如果不是，则患者代码上缺少索引。

Answer 2

获取您的原始查询并在其上使用PIVOT：

SELECT *
FROM   (
  SELECT f.FALLID AS ID,
         f.FALLNR AS Fallnummer,
         DECODE(e.SCOREID, 12, 'SAPS', 13, 'TISS', 'X') AS Score,
         TO_CHAR(e.VON, 'DD.MM.YYYY') AS Datum,
         scp.BEZEICHNUNG,
         s.PUNKTE
  FROM   FALL f
         INNER JOIN OS_MUP.EINSTUFUNG e
         ON ( e.FALLID = f.FALLID )
         INNER JOIN OS_MUP.EINSTUFUNGSRANG einr
         ON ( einr.EINSTUFUNGID = e.EINSTUFUNGID )
         INNER JOIN OS_MUP.SCORERANG s
         ON ( s.SCORERANGID = einr.SCORERANGID )
         JOIN OS_MUP.SCOREPOSITION scp
         ON ( scp.SCOREPOSITIONID = s.SCOREPOSITIONID )
  WHERE  f.FALLNR = 1234567
)
PIVOT (
  MAX( Punkte ) FOR BEZEICHNUNG IN (
    'Age' AS Age,
    'Temperature' AS Temperature,
    'Alter' AS Alter,
    'Herzfrequenz' AS Herzfrequenz,
    'RR Syst.' AS "RR Syst.",
    'Körpertemp' AS "Körpertemp.",
    'PaO2/FiO2' AS "PaO2/FiO2"
    -- ...
  )
)
ORDER BY
       SCOREID,
       VON