Oracle 19c Automatic Indexing: How Many Executions Does It Take? (One Shot)

One of the first questions I asked when playing with the new Oracle Database 19c Automatic Indexing feature was how many executions of an SQL does it take for a new index to be considered?

To find out, I create the following table:

SQL> create table bowie_one (id number constraint bowie_one_pk primary key, code number, name varchar2(42));

Table created.

SQL> insert into bowie_one select rownum, mod(rownum, 1000000)+1, 'David Bowie' from dual connect by level

SQL> commit;

Commit complete.

SQL> exec dbms_stats.gather_table_stats(ownname=>null, tabname=>'BOWIE_ONE');

PL/SQL procedure successfully completed.

I then ran the following query just once and checked to see if the Automatic Indexing task would pick this execution up and consider building a new index:

SQL> select * from bowie_one where code=42;

10 rows selected.

Execution Plan

------------------------------------------------------------------------------------------
| Id  | Operation                    | Name      | Rows  | Bytes | Cost (%CPU)| Time     |
------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |           |    10 |   230 |  6208   (7)| 00:00:01 |
|   1 |  PX COORDINATOR              |           |       |       |            |          |
|   2 |   PX SEND QC (RANDOM)        | :TQ10000  |    10 |   230 |  6208   (7)| 00:00:01 |
|   3 |    PX BLOCK ITERATOR         |           |    10 |   230 |  6208   (7)| 00:00:01 |
|*  4 |     TABLE ACCESS STORAGE FULL| BOWIE_ONE |    10 |   230 |  6208   (7)| 00:00:01 |
------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):

---------------------------------------------------

4 - storage("CODE"=42)
    filter("CODE"=42)

Statistics
----------------------------------------------------------

   12  recursive calls
    0  db block gets
39000  consistent gets
    0  physical reads
  132  redo size
  867  bytes sent via SQL*Net to client
  588  bytes received via SQL*Net from client
    2  SQL*Net roundtrips to/from client
    0  sorts (memory)
    0  sorts (disk)
   10  rows processed

The following Automatic Indexing report detailed the following:

SQL> select dbms_auto_index.report_last_activity() report from dual;

REPORT
--------------------------------------------------------------------------------
GENERAL INFORMATION
-------------------------------------------------------------------------------
Activity start               : 26-JUN-2019 13:03:30
Activity end                 : 26-JUN-2019 21:13:06
Executions completed         : 24
Executions interrupted       : 0
Executions with fatal error  : 0
-------------------------------------------------------------------------------
SUMMARY (AUTO INDEXES)
-------------------------------------------------------------------------------
Index candidates                             : 1
Indexes created (visible / invisible)        : 1 (1 / 0)
Space used (visible / invisible)             : 184.55 MB (184.55 MB / 0 B)
Indexes dropped                              : 0
SQL statements verified                      : 3
SQL statements improved (improvement factor) : 1 (19500x)
SQL plan baselines created                   : 0
Overall improvement factor                   : 6.9x
-------------------------------------------------------------------------------
SUMMARY (MANUAL INDEXES)
-------------------------------------------------------------------------------

Unused indexes    : 0
Space used        : 0 B
Unusable indexes  : 0

So an index was indeed created. Later in the report:

INDEX DETAILS

-------------------------------------------------------------------------------
The following indexes were created:
-------------------------------------------------------------------------

-------------------------------------------------------------------------
| Owner | Table     | Index                | Key  | Type   | Properties |
-------------------------------------------------------------------------
| BOWIE | BOWIE_ONE | SYS_AI_5tabfu6wtkbdh | CODE | B-TREE | NONE       |
-------------------------------------------------------------------------
-------------------------------------------------------------------------------

VERIFICATION DETAILS

-------------------------------------------------------------------------------The performance of the following statements improved:
-------------------------------------------------------------------------------

Parsing Schema Name  : BOWIE
SQL ID               : 9n89axkwrvw4b
SQL Text             : select * from bowie_one where code=42
Improvement Factor   : 19500x

Execution Statistics:
-----------------------------

                    Original Plan                 Auto Index Plan
                    ----------------------------  ----------------------------
Elapsed Time (s):   198342                        961
CPU Time (s):       187768                        1112
Buffer Gets:        39000                         13
Optimizer Cost:     6208                          14
Disk Reads:         0                             2
Direct Writes:      0                             0
Rows Processed:     10                            10
Executions:         1                             1

So the above details that an index on the CODE column of the BOWIE_ONE table was indeed created after just 1 execution.

For those wondering, yes the Elaspsed and CPU times are actually in Microseconds (1 millionth of a second) and not in seconds as stated…

The final section of the report details:

PLANS SECTION
---------------------------------------------------------------------------------------------
- Original
-----------------------------

Plan Hash Value  : 227986582
------------------------------------------------------------------------------------
| Id | Operation                      | Name      | Rows | Bytes | Cost | Time     |
------------------------------------------------------------------------------------
|  0 | SELECT STATEMENT               |           |      |       | 6208 |          |
|  1 |  PX COORDINATOR                |           |      |       |      |          |
|  2 |    PX SEND QC (RANDOM)         | :TQ10000  |   10 |   230 | 6208 | 00:00:01 |
|  3 |     PX BLOCK ITERATOR          |           |   10 |   230 | 6208 | 00:00:01 |
|  4 |      TABLE ACCESS STORAGE FULL | BOWIE_ONE |   10 |   230 | 6208 | 00:00:01 |
------------------------------------------------------------------------------------

Notes
-----

- dop_op_reason = scan of object BOWIE.BOWIE_ONE
- dop = 2
- px_in_memory_imc = no
- px_in_memory = no
- With Auto Indexes
-----------------------------

Plan Hash Value  : 2734060610
-------------------------------------------------------------------------------------------------------
| Id  | Operation                             | Name                 | Rows | Bytes | Cost | Time     |
-------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT                      |                      |  10   |  230 |   14 | 00:00:01 |
|   1 |   TABLE ACCESS BY INDEX ROWID BATCHED | BOWIE_ONE            |  10   |  230 |   14 | 00:00:01 |
| * 2 |    INDEX RANGE SCAN                   | SYS_AI_5tabfu6wtkbdh |  10   |      |    3 | 00:00:01 |
-------------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
------------------------------------------

* 2 - access("CODE"=42)

Notes
-----

- Dynamic sampling used for this statement ( level = 11 )

It details that indeed, a new plan using the newly Automatic Index would  be substantially more efficient.

If we look at details of the new Automatic Index:

SQL> select index_name, auto, constraint_index, visibility, compression, status, num_rows, leaf_blocks, clustering_factor  from user_indexes where table_name='BOWIE_ONE';

INDEX_NAME             AUT CON VISIBILIT COMPRESSION   STATUS     NUM_ROWS LEAF_BLOCKS CLUSTERING_FACTOR
---------------------- --- --- --------- ------------- -------- ---------- ----------- -----------------
BOWIE_ONE_PK           NO  YES VISIBLE   DISABLED      VALID      10000000       19642             57523
SYS_AI_5tabfu6wtkbdh   YES NO  VISIBLE   DISABLED      VALID      10000000       22285          10000000

SQL> select index_name, column_name, column_position
from user_ind_columns where table_name='BOWIE_ONE' order by index_name, column_position;

INDEX_NAME             COLUMN_NAME     COLUMN_POSITION
---------------------- --------------- ---------------
BOWIE_ONE_PK           ID                            1
SYS_AI_5tabfu6wtkbdh   CODE                          1

The newly created Automatic Index is both Valid and Visible and so can be used globally within the database.

If I now re-run the original query:

SQL> select * from bowie_one where code=42;

10 rows selected.

Execution Plan
--------------------------------------------------------------------------------------------------------------
| Id  | Operation                             | Name                 | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT                      |                      |    10 |   230 |    13   (0)| 00:00:01 |
|   1 |  PX COORDINATOR                       |                      |       |       |            |          |
|   2 |   PX SEND QC (RANDOM)                 | :TQ10001             |    10 |   230 |    13   (0)| 00:00:01 |
|   3 |    TABLE ACCESS BY INDEX ROWID BATCHED| BOWIE_ONE            |    10 |   230 |    13   (0)| 00:00:01 |
|   4 |     BUFFER SORT                       |                      |       |       |            |          |
|   5 |      PX RECEIVE                       |                      |    10 |       |     3   (0)| 00:00:01 |
|   6 |       PX SEND HASH (BLOCK ADDRESS)    | :TQ10000             |    10 |       |     3   (0)| 00:00:01 |
|   7 |        PX SELECTOR                    |                      |       |       |            |          |
|*  8 |           INDEX RANGE SCAN            | SYS_AI_5tabfu6wtkbdh |    10 |       |     3   (0)| 00:00:01 |
--------------------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

8 - access("CODE"=42)

Statistics
---------------------------------------------------------

 12  recursive calls
  0  db block gets
 13  consistent gets
  0  physical reads
  0  redo size
867  bytes sent via SQL*Net to client
588  bytes received via SQL*Net from client
  2  SQL*Net roundtrips to/from client
  2  sorts (memory)
  0  sorts (disk)
 10  rows processed

The CBO now uses the newly created Automatic Index.

So it only potentially takes just the one execution of an SQL statement for an Automatic Index to be created.

Therefore some caution needs to be exercised in environments where there may be a very large number of ad-hoc queries where specific indexes may not be necessary for once only executed predicate combinations.

That said, the Automatic Indexing process is highly efficient in building only the bare minimum of column indexed combinations to cater for all known SQL predicates.

More on this in a future post.