Oracle 中临时数据的性能注意事项

I'm evaluating various options to run a bunch of high-performing queries against a single temporary data set in Oracle. In T-SQL, I'd probably use in-memory temporary tables, but Oracle doesn't have an exact equivalent of this feature.

我目前看到这些选项:

CREATE GLOBAL TEMPORARY TABLE test_temp_t (
  n NUMBER(10), 
  s VARCHAR2(10)
) ON COMMIT DELETE ROWS; -- Other configurations are possible, too

DECLARE
  t test_t;
  n NUMBER(10);
BEGIN

  -- Replace this with the actual temporary data set generation
  INSERT INTO test_temp_t
  SELECT MOD(level, 10), '' || MOD(level, 12)
  FROM dual
  CONNECT BY level < 1000000;

  -- Replace this example query with more interesting statistics
  SELECT COUNT(DISTINCT t.n)
  INTO n 
  FROM test_temp_t t;

  DBMS_OUTPUT.PUT_LINE(n);
END;

计划:

----------------------------------------------------
| Id  | Operation            | A-Rows |   A-Time   |
----------------------------------------------------
|   0 | SELECT STATEMENT     |      1 |00:00:00.27 |
|   1 |  SORT AGGREGATE      |      1 |00:00:00.27 |
|   2 |   VIEW               |     10 |00:00:00.27 |
|   3 |    HASH GROUP BY     |     10 |00:00:00.27 |
|   4 |     TABLE ACCESS FULL|    999K|00:00:00.11 |
----------------------------------------------------

2.PL/SQL 表类型变量的取消嵌套

CREATE TYPE test_o AS OBJECT (n NUMBER(10), s VARCHAR2(10));
CREATE TYPE test_t AS TABLE OF test_o;

DECLARE
  t test_t;
  n NUMBER(10);
BEGIN

  -- Replace this with the actual temporary data set generation
  SELECT test_o(MOD(level, 10), '' || MOD(level, 12))
  BULK COLLECT INTO t
  FROM dual
  CONNECT BY level < 1000000;

  -- Replace this example query with more interesting statistics
  SELECT COUNT(DISTINCT n)
  INTO n 
  FROM TABLE(t) t;

  DBMS_OUTPUT.PUT_LINE(n);
END;

计划:

------------------------------------------------------------------
| Id  | Operation                          | A-Rows |   A-Time   |
------------------------------------------------------------------
|   0 | SELECT STATEMENT                   |      1 |00:00:00.68 |
|   1 |  SORT GROUP BY                     |      1 |00:00:00.68 |
|   2 |   COLLECTION ITERATOR PICKLER FETCH|    999K|00:00:00.22 |
------------------------------------------------------------------

3.物化视图

对于这个用例，我将它们排除在外，因为有问题的临时数据集相当复杂，更新物化视图的影响太大.

Type  Name                              Run1 (temp) Run2 (PLSQL)         Diff
----- -------------------------------- ------------ ------------ ------------
...
STAT  physical read bytes                    81,920            0      -81,920
STAT  physical read total bytes              81,920            0      -81,920
LATCH cache buffers chains                  104,663          462     -104,201
STAT  session uga memory                    445,488      681,016      235,528
STAT  KTFB alloc space (block)            2,097,152            0   -2,097,152
STAT  undo change vector size             2,350,188            0   -2,350,188
STAT  redo size                           2,804,516            0   -2,804,516
STAT  temp space allocated (bytes)       12,582,912            0  -12,582,912
STAT  table scan rows gotten             15,499,845            0  -15,499,845
STAT  session pga memory                    196,608   19,857,408   19,660,800
STAT  logical read bytes from cache     299,958,272            0 -299,958,272

select name, value
from v$sysstat
where name in ('db block gets from cache', 'consistent gets from cache', 
'physical reads cache');