获取前 10 个未使用的手册序列号

我想从某个范围中查找前 10 个未使用的手动序列号。

请在下面找到我的查询：

select X1.* From
    (Select Rownum seq_number From Dual Connect By Rownum <= 
         (Select LPAD(9,(UTC.DATA_PRECISION - UTC.DATA_SCALE),9) 
          From User_Tab_Columns UTC 
          where UTC.Table_Name = 'Table_Name' And UTC.Column_Name = 'seq_number')) X1,
Table_Name X2
Where X1.seq_number = X2.seq_number (+)
  And X2.Rowid is Null
  And Rownum <= 10

尽管它提供了所需的输出，但我担心[如果有的话]造成的负载，因为我们每天将多次使用此查询。

请告知是否有办法优化此查询。

附加信息： 在 Table_Name T2 上，在 (seq_number) 上定义了唯一索引

工作示例：

create table TEMP_TABLE_NAME ( seq_number number(6))

insert into TEMP_TABLE_NAME 
select distinct trunc(dbms_random.VALUE(1,5000)) seq_number 
from dual
connect by rownum <= 1000


create unique index TEMP_TABLE_NAME_IDX on TEMP_TABLE_NAME(seq_Number)


SELECT T1.*
  FROM (    SELECT ROWNUM seq_number
              FROM DUAL
        CONNECT BY ROWNUM <=
                      (SELECT LPAD (9,(UTC.DATA_PRECISION - UTC.DATA_SCALE),9)
                         FROM User_Tab_Columns UTC
                        WHERE     UTC.Table_Name = 'TEMP_TABLE_NAME'
                              AND UTC.Column_Name = 'SEQ_NUMBER')) T1,
       TEMP_TABLE_NAME T2
 WHERE     T1.seq_number = T2.seq_number(+)
       AND T2.ROWID IS NULL
       AND ROWNUM <= 10

对我来说，我的查询给出了以下输出。表中随机创建的数字包括 7 和 8，因此它们被忽略。重点是获取前 10 个未使用的数字。

1
2
3
4
5
6
9
10
11
12

First我将替换这个复杂的子查询：

Select Rownum seq_number From Dual Connect By Rownum <= 
         (Select LPAD(9,(UTC.DATA_PRECISION - UTC.DATA_SCALE),9) 
          From User_Tab_Columns UTC 
          where UTC.Table_Name = 'Table_Name' And UTC.Column_Name = 'seq_number')

与这个：

Select Rownum As seq_number From Dual 
Connect By Rownum <= (Select max( seq_number ) + 10 From TEMP_TABLE_NAME ) 

甚至用一个简单的常数：

Select Rownum As seq_number From Dual Connect By Rownum <= 1000000

坦率地说，您的子查询不适用于非常基本的情况：

create table TEMP_TABLE_NAME(
  seq_number NUMBER
);

SELECT LPAD (9,(UTC.DATA_PRECISION - UTC.DATA_SCALE),9) as x , 
       UTC.DATA_PRECISION, UTC.DATA_SCALE, UTC.COLUMN_NAME
FROM User_Tab_Columns UTC
WHERE     UTC.Table_Name = 'TEMP_TABLE_NAME'
  AND UTC.Column_Name = 'SEQ_NUMBER'
;

X        DATA_PRECISION DATA_SCALE COLUMN_NAME
-------- -------------- ---------- -----------
  (null)         (null)     (null) SEQ_NUMBER

还有第二种情况：

create table TEMP_TABLE_NAME(
  seq_number NUMBER(15,0)
);

在这种情况下，子查询尝试生成 999999999999999 行，这很快就会导致内存不足错误

SELECT count(*) FROM (
 SELECT ROWNUM seq_number
              FROM DUAL
        CONNECT BY ROWNUM <=
                      (SELECT LPAD (9,(UTC.DATA_PRECISION - UTC.DATA_SCALE),9)
                         FROM User_Tab_Columns UTC
                        WHERE     UTC.Table_Name = 'TEMP_TABLE_NAME'
                              AND UTC.Column_Name = 'SEQ_NUMBER')
);

ORA-30009: Not enough memory for CONNECT BY operation
30009. 0000 -  "Not enough memory for %s operation"
*Cause:    The memory size was not sufficient to process all the levels of the
           hierarchy specified by the query.
*Action:   In WORKAREA_SIZE_POLICY=AUTO mode, set PGA_AGGREGATE_TARGET to
           a reasonably larger value.
           Or, in WORKAREA_SIZE_POLICY=MANUAL mode, set SORT_AREA_SIZE to a
           reasonably larger value.

其次，您的查询不是确定性的！
它强烈依赖于物理表结构，并且不使用强加正确的顺序ORDER BY clause.
记住->维基百科 - 排序依据 https://en.wikipedia.org/wiki/Order_by

ORDER BY是对结果集中的行进行排序的唯一方法。没有 该子句中，关系数据库系统可以返回任意行中的行 命令。如果需要订购，则必须提供 ORDER BY 应用程序发送的 SELECT 语句。

考虑这个测试用例：

create table TEMP_TABLE_NAME 
as SELECT * FROM (
    select rownum as seq_number , t.*
    from ALL_OBJECTS t
    cross join ( select * from dual connect by level <= 10)
    where rownum <= 100000
)
ORDER BY DBMS_RANDOM.Value;
create unique index TEMP_TABLE_NAME_IDX on TEMP_TABLE_NAME(seq_Number);

select count(*) from TEMP_TABLE_NAME;
  COUNT(*)
----------
    100000

DELETE FROM TEMP_TABLE_NAME
WHERE seq_number between 10000 and 10002
  OR seq_number between 20000 and 20002
  OR seq_number between 30000 and 30002
  OR seq_number between 40000 and 40002
  OR seq_number between 50000 and 50002
  OR seq_number between 60000 and 60002
  ;

如果索引存在，那么结果就OK：

SELECT T1.*
  FROM (    SELECT ROWNUM seq_number
              FROM DUAL
        CONNECT BY ROWNUM <= 1000000
) T1,
       TEMP_TABLE_NAME T2
 WHERE     T1.seq_number = T2.seq_number(+)
       AND T2.ROWID IS NULL
       AND ROWNUM <= 10
;

SEQ_NUMBER
----------
     10000
     10001
     10002
     20000
     20001
     20002
     30000
     30001
     30002
     40000

但是，当有一天有人删除索引，或者优化器由于某种原因决定不使用该索引时，会发生什么情况？
根据定义：如果没有 ORDER BY，关系数据库系统可能会以任何顺序返回行。我使用提示模拟这些情况：

SELECT /*+ NO_INDEX(T2) */ T1.*
  FROM (    SELECT ROWNUM seq_number
              FROM DUAL
        CONNECT BY ROWNUM <= 1000000
) T1,
       TEMP_TABLE_NAME T2
 WHERE     T1.seq_number = T2.seq_number(+)
       AND T2.ROWID IS NULL
       AND ROWNUM <= 10
;

SEQ_NUMBER
----------
    213856
    910281
    668862
    412743
    295487
    214762
    788486
    346216
    777734
    806457

下面的查询强制使用正确的顺序ORDER BY子句并给出可重现的结果，无论是否存在正确的索引。
我正在使用推荐的 ANSI SQL LEFT JOIN 子句，而不是过时的WHERE .... (+) syntax.

SELECT  * FROM (
    SELECT /*+ NO_INDEX(T2) */ T1.*
      FROM (    SELECT ROWNUM seq_number
                  FROM DUAL
            CONNECT BY ROWNUM <= 1000000
    ) T1 
    LEFT JOIN TEMP_TABLE_NAME T2
    ON T1.seq_number = T2.seq_number
    WHERE T2.ROWID IS NULL
    ORDER BY T1.seq_number
)
WHERE ROWNUM <= 10

表现
检查性能的最简单方法是进行测试 - 运行查询 10-100 次并测量时间：

SET TIMING ON;
DECLARE
   x NUMBER;
BEGIN
   FOR i IN 1..10 LOOP
      SELECT sum( seq_number ) INTO x
      FROM (
           SELECT  * FROM (
            SELECT T1.*
              FROM (    SELECT ROWNUM seq_number
                          FROM DUAL
                    CONNECT BY ROWNUM <= 1000000
            ) T1 
            LEFT JOIN TEMP_TABLE_NAME T2
            ON T1.seq_number = T2.seq_number
            WHERE T2.ROWID IS NULL
            ORDER BY T1.seq_number
            )
            WHERE ROWNUM <= 10
        );
    END LOOP;
END;
/

PL/SQL procedure successfully completed.

Elapsed: 00:00:11.750

10 次 - 11.75 秒，因此一次查询需要 1.2 秒。

下一个版本有限制CONNECT BY使用子查询：

SET TIMING ON;
DECLARE
   x NUMBER;
BEGIN
   FOR i IN 1..10 LOOP
      SELECT sum( seq_number ) INTO x
      FROM (
           SELECT  * FROM (
            SELECT T1.*
              FROM (    SELECT ROWNUM seq_number
                          FROM DUAL
                    CONNECT BY ROWNUM <= (Select max( seq_number ) + 10 From TEMP_TABLE_NAME ) 
            ) T1 
            LEFT JOIN TEMP_TABLE_NAME T2
            ON T1.seq_number = T2.seq_number
            WHERE T2.ROWID IS NULL
            ORDER BY T1.seq_number
            )
            WHERE ROWNUM <= 10
        );
    END LOOP;
END;
/
PL/SQL procedure successfully completed.

Elapsed: 00:00:00.986

好多了 - 只需 100 毫秒。
由此得出的结论是，CONNECT BY部分是成本最高的。

另一种尝试是使用预先生成的数字序列高达 1 mln（某种物化视图）的表来代替CONNECT BY每次在内存中动态生成数字的子查询：

create table seq(
   seq_number int primary key
)
ORGANIZATION INDEX ;

INSERT INTO seq 
SELECT level FROM dual
CONNECT BY LEVEL <= 1000000;

SET TIMING ON;
DECLARE
   x NUMBER;
BEGIN
   FOR i IN 1..10 LOOP
      SELECT sum( seq_number ) INTO x
      FROM (
           SELECT  * FROM (
            SELECT T1.*
            FROM seq T1 
            LEFT JOIN TEMP_TABLE_NAME T2
            ON T1.seq_number = T2.seq_number
            WHERE T2.ROWID IS NULL
            ORDER BY T1.seq_number
            )
            WHERE ROWNUM <= 10
        );
    END LOOP;
END;
/

PL/SQL procedure successfully completed.

Elapsed: 00:00:00.398

这是最快的 - 仅 40 毫秒

第一个 1200 毫秒，最后一个 40 毫秒 - 快了 30 倍 (3000 %)。