oracle排序操作

1. 查询排序最多的SQL语句：

   WITH sql_workarea AS
    (SELECT sql_id || '_' || child_number sql_id_child,
            operation_type operation,
            last_execution last_exec,
            round(active_time / 1000000, 2) seconds,
            optimal_executions || '/' || multipasses_executions olm,
            '' || substr(sql_text, 1, 155) sql_text,
            rank() over(ORDER BY active_time DESC) ranking
       FROM v$sql_workarea
       JOIN v$sql
      USING (sql_id, child_number))
   SELECT sql_id_child "SQL ID-CHILD",
          seconds,
          operation,
          last_exec,
          olm          "O/1/M",
          sql_text
     FROM sql_workarea
    WHERE ranking <= 10
    ORDER BY ranking;
   

2. 10033跟踪排序：

   alter session set tracefile_identifier=e10033;

        alter session set events '10033 trace name context forever,level 1';
---- Sort Statistics ------------------------------
Initial runs                              14
Number of merges                          1
Input records                             55500
Output records                            55500
Disk blocks 1st pass                      1467
Total disk blocks used                    1451
Total number of comparisons performed     699074
  Comparisons performed by in-memory sort 485849
  Comparisons performed during merge      213212
  Comparisons while searching for key in-memory 13
Number of seeks in final run              55500
Temp segments allocated                   1
Extents allocated                         12
Uses version 2 sort
Uses asynchronous IO
    ---- Run Directory Statistics ----
Run directory block reads (buffer cache)  15
Block pins (for run directory)            1
Block repins (for run directory)          14
Maximum input run size (in blocks)        109
Minimum input run size (in blocks)        32
Average input run size (in blocks)        104
    ---- Direct Write Statistics -----
Write slot size                           49152
Write slots used during in-memory sort    2
Number of direct writes                   247
Num blocks written (with direct write)    1449
Block pins (for sort records)             1449
Waits for async writes                    199
    ---- Direct Read Statistics ------
Size of read slots for output             32768
Number of read slots for output           32
Number of direct sync reads               30
Number of blocks read synchronously       95
Number of direct async reads              343
Number of blocks read asynchronously      1354

使用索引来规避排序

如果在order by字句中的部分或者全部列上存在索引，oracle有可能使用索引来按照要求的顺序获取记录，因此也避免了排序操作。

假如索引是出现在与orde by字句里的列相同的列上，oracle可以直接从索引中按照索引排序的顺序读取记录，然而，按键的顺序读取记录需要一块接一块地全扫描索引叶子块。虽然快速全扫描比全索引扫描高校得多，但是快速全扫描无法按索引顺序返回记录，因此也不能用来避免排序操作。

SQL> select * from customers order by cust_last_name,cust_first_name,cust_year_of_birth;

55500 rows selected.
Execution Plan
----------------------------------------------------------
Plan hash value: 2792773903

----------------------------------------------------------------------------------------
| Id  | Operation    | Name      | Rows  | Bytes |TempSpc| Cost (%CPU)| Time     |
----------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |        | 55500 |  9810K|       |  2609  (1)| 00:00:02 |
|   1 |  SORT ORDER BY    |        | 55500 |  9810K|    12M|  2609  (1)| 00:00:02 |
|   2 |   TABLE ACCESS FULL| CUSTOMERS | 55500 |  9810K|       |   405  (1)| 00:00:01 |
----------------------------------------------------------------------------------------
Statistics
----------------------------------------------------------
  12  recursive calls
  15  db block gets
       1456  consistent gets
       2903  physical reads
   0  redo size
    6366362  bytes sent via SQL*Net to client
      41213  bytes received via SQL*Net from client
       3701  SQL*Net roundtrips to/from client
   0  sorts (memory)
   1  sorts (disk)
      55500  rows processed

建索引后：

SQL> create index cust_namedob_i on customers(cust_last_name,cust_first_name,cust_year_of_birth);

Index created.

SQL> select /*+ index(customers,cust_namedob_i) */ * from customers order by cust_last_name,cust_first_name,cust_year_of_birth;

55500 rows selected.

Execution Plan
----------------------------------------------------------
Plan hash value: 1819843466

----------------------------------------------------------------------------------------------
| Id  | Operation      | Name      | Rows  | Bytes | Cost (%CPU)| Time     |
----------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT     |       | 55500 | 9810K| 20550   (1)| 00:00:15 |
|   1 |  TABLE ACCESS BY INDEX ROWID| CUSTOMERS      | 55500 | 9810K| 20550   (1)| 00:00:15 |
|   2 |   INDEX FULL SCAN     | CUST_NAMEDOB_I | 55500 |      |  225   (0)| 00:00:01 |
----------------------------------------------------------------------------------------------
Statistics
----------------------------------------------------------
   1  recursive calls
   0  db block gets
      26557  consistent gets
       1708  physical reads
   0  redo size
    6366312  bytes sent via SQL*Net to client
      41213  bytes received via SQL*Net from client
       3701  SQL*Net roundtrips to/from client
   0  sorts (memory)
   0  sorts (disk)
      55500  rows processed

虽然使用索引可能就不再需要排序了，但是同时读取索引和表块，以及按块顺次读取这种并不高效的扫描方式所带来的开销，比使用全表扫描读取表块的方式要欠佳很多，通常，这意味为了避免排序而使用索引，实际上会导致更差的性能。然而使用索引在检索第一行记录时速度更快，因为一旦需要的记录被检索到，它会立即返回。相比之下排序的方法要求在任一记录返回之前，全部记录都必须被检索出来并完成排序。因此，在优化器目标为FIRST_ROWS_N时，优化器倾向于使用索引，而在目标是ALL_ROWS时，则会使用全表扫描。

  另一个基于索引的获取比先扫描再获取要更优异的场景是当内存极其有限时。如果可用于排序的内存是受限的，读写临时段所需要IO将超过索引和和表扫描所包含的额外的IO开销。当然如果能够分配更多的内存，它的表现会好很多的，但是如果这是不可能的，你或许应该使用INDEX提示来避免排序。

聚合操作

聚合炒作（如SUM和AVG）必须处理输入的数据每一行记录，因此，它们通常和全表扫描联系在一起。

SQL> select sum(quantity_sold) from sales;
Execution Plan
----------------------------------------------------------
Plan hash value: 3519235612

----------------------------------------------------------------------------------------------
| Id  | Operation      | Name  | Rows  | Bytes | Cost (%CPU)| Time     | Pstart| Pstop |
----------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT     |      |    1 |    3 |  525   (2)| 00:00:01 |      |      |
|   1 |  SORT AGGREGATE      |      |    1 |    3 |    |      |      |      |
|   2 |   PARTITION RANGE ALL|      |  918K| 2691K|  525   (2)| 00:00:01 |    1 |   28 |
|   3 |    TABLE ACCESS FULL | SALES |  918K| 2691K|  525   (2)| 00:00:01 |    1 |   28 |
----------------------------------------------------------------------------------------------
Statistics
----------------------------------------------------------
       2429  recursive calls
   2  db block gets
       5371  consistent gets
       1714  physical reads
   0  redo size
 538  bytes sent via SQL*Net to client
 524  bytes received via SQL*Net from client
   2  SQL*Net roundtrips to/from client
 183  sorts (memory)
   0  sorts (disk)
   1  rows processed

SQL> select /*+ index(sales,index_sl) */ sum(quantity_sold) from sales;

SUM(QUANTITY_SOLD)
------------------
     918843

Execution Plan
----------------------------------------------------------
Plan hash value: 3788238680

-----------------------------------------------------------------------------
| Id  | Operation  | Name     | Rows  | Bytes | Cost (%CPU)| Time     |
-----------------------------------------------------------------------------
|   0 | SELECT STATEMENT |     |   1 |   3 |  2316   (1)| 00:00:02 |
|   1 |  SORT AGGREGATE  |     |   1 |   3 |   |     |
|   2 |   INDEX FULL SCAN| INDEX_SL | 918K|  2691K|  2316   (1)| 00:00:02 |
-----------------------------------------------------------------------------

Statistics
----------------------------------------------------------
   1  recursive calls
   0  db block gets
       2311  consistent gets
       2314  physical reads
   0  redo size
 538  bytes sent via SQL*Net to client
 524  bytes received via SQL*Net from client
   2  SQL*Net roundtrips to/from client
   0  sorts (memory)
   0  sorts (disk)
   1  rows processed

最大值和最小值，

与大多数其他的聚合操作不同，如果在相关列存在索引，MAX和MIN操作并不需要读取每一行记录。如果存在B树索引，我们可以通过检查第一个或最后一个索引项来确定最大值或最小值，这仅需要3-5个逻辑读的开销：

SQL> select max(amount_sold) from sales;

MAX(AMOUNT_SOLD)
----------------
  1782.72

Execution Plan
----------------------------------------------------------
Plan hash value: 781264156

----------------------------------------------------------------------------------------------
| Id  | Operation     | Name      | Rows  | Bytes | Cost (%CPU)| Time     |
----------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT    |       |    1 |    5 |    3   (0)| 00:00:01 |
|   1 |  SORT AGGREGATE     |       |    1 |    5 |    |      |
|   2 |   INDEX FULL SCAN (MIN/MAX)| AMOUNT_SOLD_IDX |    1 |    5 |    3   (0)| 00:00:01 |
----------------------------------------------------------------------------------------------
Statistics
----------------------------------------------------------
   1  recursive calls
   0  db block gets
   3  consistent gets
   8  physical reads
   0  redo size
 536  bytes sent via SQL*Net to client
 524  bytes received via SQL*Net from client
   2  SQL*Net roundtrips to/from client
   0  sorts (memory)
   0  sorts (disk)
   1  rows processed

同时找出最大值和最小值，

SQL> select max(amount_sold),min(amount_sold) from sales;

MAX(AMOUNT_SOLD) MIN(AMOUNT_SOLD)
---------------- ----------------
  1782.72       6.4

Execution Plan
----------------------------------------------------------
Plan hash value: 3519235612

----------------------------------------------------------------------------------------------
| Id  | Operation      | Name  | Rows  | Bytes | Cost (%CPU)| Time     | Pstart| Pstop |
----------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT     |      |    1 |    5 |  525   (2)| 00:00:01 |      |      |
|   1 |  SORT AGGREGATE      |      |    1 |    5 |    |      |      |      |
|   2 |   PARTITION RANGE ALL|      |  918K| 4486K|  525   (2)| 00:00:01 |    1 |   28 |
|   3 |    TABLE ACCESS FULL | SALES |  918K| 4486K|  525   (2)| 00:00:01 |    1 |   28 |
----------------------------------------------------------------------------------------------
Statistics
----------------------------------------------------------
   1  recursive calls
   0  db block gets
       1635  consistent gets
       1619  physical reads
   0  redo size
 618  bytes sent via SQL*Net to client
 524  bytes received via SQL*Net from client
   2  SQL*Net roundtrips to/from client
   0  sorts (memory)
   0  sorts (disk)
   1  rows processed

实际上分别提交MAX和MIN查询然后将结果合并到一起是一种更好的方法：

SELECT max_sold, min_sold
  FROM (SELECT MAX(amount_sold) max_sold FROM sales) maxt,
  2    3         (SELECT MIN(amount_sold) min_sold FROM sales) mint;

  MAX_SOLD   MIN_SOLD
---------- ----------
   1782.72   6.4

Execution Plan
----------------------------------------------------------
Plan hash value: 3650580342

------------------------------------------------------------------------------------------------
| Id  | Operation       | Name        | Rows  | Bytes | Cost (%CPU)| Time     |
------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT      |         |     1 |    26 |     6  (0)| 00:00:01 |
|   1 |  NESTED LOOPS       |         |     1 |    26 |     6  (0)| 00:00:01 |
|   2 |   VIEW        |         |     1 |    13 |     3  (0)| 00:00:01 |
|   3 |    SORT AGGREGATE      |         |     1 |     5 |     |        |
|   4 |     INDEX FULL SCAN (MIN/MAX)| AMOUNT_SOLD_IDX |     1 |     5 |     3  (0)| 00:00:01 |
|   5 |   VIEW        |         |     1 |    13 |     3  (0)| 00:00:01 |
|   6 |    SORT AGGREGATE      |         |     1 |     5 |     |        |
|   7 |     INDEX FULL SCAN (MIN/MAX)| AMOUNT_SOLD_IDX |     1 |     5 |     3  (0)| 00:00:01 |
------------------------------------------------------------------------------------------------

Statistics
----------------------------------------------------------
   1  recursive calls
   0  db block gets
   6  consistent gets
   5  physical reads
   0  redo size
 602  bytes sent via SQL*Net to client
 524  bytes received via SQL*Net from client
   2  SQL*Net roundtrips to/from client
   0  sorts (memory)
   0  sorts (disk)
   1  rows processed

 前N 查询

如何获取一个表的前10行记录，

错误写法：

SQL> SELECT * FROM sales WHERE rownum <= 10 ORDER BY amount_sold DESC;

   PROD_ID    CUST_ID TIME_ID    CHANNEL_ID   PROMO_ID QUANTITY_SOLD AMOUNT_SOLD
---------- ---------- ------------------- ---------- ---------- ------------- -----------
 13   987 1998-01-10 00:00:00    3     999      1   1232.16
 13  1660 1998-01-10 00:00:00    3     999      1   1232.16
 13  1762 1998-01-10 00:00:00    3     999      1   1232.16
 13  1843 1998-01-10 00:00:00    3     999      1   1232.16
 13  4663 1998-01-10 00:00:00    3     999      1   1232.16
 13  2273 1998-01-10 00:00:00    3     999      1   1232.16
 13  2380 1998-01-10 00:00:00    3     999      1   1232.16
 13  2683 1998-01-10 00:00:00    3     999      1   1232.16
 13  2865 1998-01-10 00:00:00    3     999      1   1232.16
 13  1948 1998-01-10 00:00:00    3     999      1   1232.16

10 rows selected.

这是因为对where的处理会先于order by。因此这个查询将获取它最先发现的10条记录，然后对它们进行排序。这样的结果不是真正的前10.

下面的查询更好：

SELECT /* top10_subquery */
 *
  FROM (SELECT cust_id, prod_id, time_id, amount_sold
          FROM sales
         ORDER BY amount_sold DESC)
 WHERE rownum <= 10;
  4    5    6 
   CUST_ID    PROD_ID TIME_ID    AMOUNT_SOLD
---------- ---------- ------------------- -----------
      3948    18 1999-04-26 00:00:00     1782.72
      4150    18 1999-06-26 00:00:00     1782.72
 40    18 1999-06-26 00:00:00     1782.72
     33724    18 1999-06-21 00:00:00     1782.72
     32863    18 1999-06-21 00:00:00     1782.72
     31364    18 1999-06-21 00:00:00     1782.72
     10864    18 1999-06-21 00:00:00     1782.72
     10620    18 1999-06-21 00:00:00     1782.72
      6490    18 1999-06-21 00:00:00     1782.72
      4788    18 1999-06-21 00:00:00     1782.72

10 rows selected.