Comments on: Adding Comments to SQL Statements Improves Performance?

By: State of Data Last Week – #32 « Dr Data's Blog

State of Data Last Week – #32 « Dr Data's Blog — Sat, 22 Jan 2011 21:43:21 +0000

[...] SQL code is not only a good practice, comments could also improve SQL performance. What?? No, really – here is a use case. It’s a trick, read on the comments if the rationale [...]

By: Charles Hooper

Charles Hooper — Mon, 17 Jan 2011 12:23:49 +0000

Kerry,

Thank you for the tip regarding including comments in hint blocks, and the suggested work-around for that problem. By the way, chapter 16 of the book is excellent – one of the best chapters in the book.

By: Kerry Osborne

Kerry Osborne — Mon, 17 Jan 2011 03:07:26 +0000

Well I think you guys have it all figured out already. Keywords in a hint can cause the valid hints to be ignored. The suggestion in the book was “Just don’t do it”. Put your comments in a a separate comment construct. Like so.

/*+ full(a) */ /* This is where the comment goes … */

Since there are about a billion keywords (some of which are not documented or well known) and since Oracle can add a new one any time they want, It’s just best not to try to mix hints and comments in the same structure. I hadn’t seen that MOS note though so that was nice verification as to what I had observed and assumed to be true.

By: Ignoring hints « Oracle Scratchpad

Ignoring hints « Oracle Scratchpad — Sun, 16 Jan 2011 22:32:19 +0000

[...] Thanks to Timur Akhmadeev for telling us about bug 8432870 (you’ll need an account on MOS for the link) in his reply to Charles Hooper’s recent post. [...]

By: Martin Preiss

Martin Preiss — Sun, 16 Jan 2011 18:39:59 +0000

with regard to the single letters I looked for elements in dba_tab_columns and dba_objects with length=1 and found some of them (‘A’, ‘D’) – but the creation of a new table with a column named ‘B’ does not change the result; and of course the signs : ; , . – are not found there.

“Oracle Database does not yet understand French” – perhaps the hints have also to be translated according to the loi toubon: http://en.wikipedia.org/wiki/Toubon_Law

By: Charles Hooper

Charles Hooper — Sun, 16 Jan 2011 17:41:52 +0000

Doc ID 826893.1 is an interesting, and very helpful document – thank you for finding that. Very likely this document is supposed to explain the behavior.

However, item number 2 below the CAUSE heading suggests that these might be SQL keywords based on other comments attached to this article:
* FAST
* PERFORMANCE
* COMMENT (OK, that is a keyword, so it is addressed by item number 2 in the article)
* :
* A
* D

Additionally, based on the comment left by nedoboi, we see that Oracle Database does not yet understand French – maybe that is a bug coming to version 12.1?

By: Timur Akhmadeev

Timur Akhmadeev — Sun, 16 Jan 2011 14:44:34 +0000

Most likely it happens due to change in behavior of the query parser described in the Doc ID 826893.1.

By: Martin Preiss

Martin Preiss — Sun, 16 Jan 2011 09:21:03 +0000

Charles, it seems that even a single letter can disable a following hint - but not every letter: [code] -- Results from 11.2.0.1 SQL> SELECT /*+ a LEADING(T1) USE_NL(T2) */ 2 T1.C1, 3 T1.C2, 4 T2.C1, 5 T2.C2 6 FROM 7 T1, 8 T2 9 WHERE 10 T1.C1=T2.C1; Abgelaufen: 00:00:00.00 Ausf³hrungsplan ---------------------------------------------------------- Plan hash value: 2959412835 ----------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time | ----------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 500K| 9765K| | 14175 (1)| 00:02:51 | |* 1 | HASH JOIN | | 500K| 9765K| 10M| 14175 (1)| 00:02:51 | | 2 | TABLE ACCESS FULL| T2 | 500K| 4882K| | 4124 (1)| 00:00:50 | | 3 | TABLE ACCESS FULL| T1 | 1000K| 9765K| | 8481 (1)| 00:01:42 | ----------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 1 - access("T1"."C1"="T2"."C1") SQL> r 1 SELECT /*+ b LEADING(T1) USE_NL(T2) */ 2 T1.C1, 3 T1.C2, 4 T2.C1, 5 T2.C2 6 FROM 7 T1, 8 T2 9 WHERE 10* T1.C1=T2.C1 Abgelaufen: 00:00:00.00 Ausf³hrungsplan ---------------------------------------------------------- Plan hash value: 2984486645 --------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | --------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 500K| 9765K| 1508K (1)| 05:01:47 | | 1 | NESTED LOOPS | | | | | | | 2 | NESTED LOOPS | | 500K| 9765K| 1508K (1)| 05:01:47 | | 3 | TABLE ACCESS FULL | T1 | 1000K| 9765K| 8481 (1)| 00:01:42 | |* 4 | INDEX UNIQUE SCAN | SYS_C0010667 | 1 | | 1 (0)| 00:00:01 | | 5 | TABLE ACCESS BY INDEX ROWID| T2 | 1 | 10 | 2 (0)| 00:00:01 | --------------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 4 - access("T1"."C1"="T2"."C1") SQL> r 1 SELECT /*+ c LEADING(T1) USE_NL(T2) */ 2 T1.C1, 3 T1.C2, 4 T2.C1, 5 T2.C2 6 FROM 7 T1, 8 T2 9 WHERE 10* T1.C1=T2.C1 Abgelaufen: 00:00:00.00 Ausf³hrungsplan ---------------------------------------------------------- Plan hash value: 2984486645 --------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | --------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 500K| 9765K| 1508K (1)| 05:01:47 | | 1 | NESTED LOOPS | | | | | | | 2 | NESTED LOOPS | | 500K| 9765K| 1508K (1)| 05:01:47 | | 3 | TABLE ACCESS FULL | T1 | 1000K| 9765K| 8481 (1)| 00:01:42 | |* 4 | INDEX UNIQUE SCAN | SYS_C0010667 | 1 | | 1 (0)| 00:00:01 | | 5 | TABLE ACCESS BY INDEX ROWID| T2 | 1 | 10 | 2 (0)| 00:00:01 | --------------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 4 - access("T1"."C1"="T2"."C1") SQL> r 1 SELECT /*+ d LEADING(T1) USE_NL(T2) */ 2 T1.C1, 3 T1.C2, 4 T2.C1, 5 T2.C2 6 FROM 7 T1, 8 T2 9 WHERE 10* T1.C1=T2.C1 Abgelaufen: 00:00:00.00 Ausf³hrungsplan ---------------------------------------------------------- Plan hash value: 2959412835 ----------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes |TempSpc| Cost (%CPU)| Time | ----------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 500K| 9765K| | 14175 (1)| 00:02:51 | |* 1 | HASH JOIN | | 500K| 9765K| 10M| 14175 (1)| 00:02:51 | | 2 | TABLE ACCESS FULL| T2 | 500K| 4882K| | 4124 (1)| 00:00:50 | | 3 | TABLE ACCESS FULL| T1 | 1000K| 9765K| | 8481 (1)| 00:01:42 | ----------------------------------------------------------------------------------- [/code] At first I began to think about vovels and consonants - but that theory died soon... The hints are also invalidated by other leading signs (,.-;) and by reserved words (or, and, table).

By: nedoboi

nedoboi — Sun, 16 Jan 2011 02:24:53 +0000

Unfortunately this technique doesn't seem to work if you comment in French -

SELECT /*+ Essayez un nouveau navigateur avec traduction automatique LEADING(T1) USE_NL(T2) */
  T1.C1,
  T1.C2,
  T2.C1,
  T2.C2
FROM
  T1,
  T2
WHERE
  T1.C1=T2.C1;

Execution Plan
----------------------------------------------------------
Plan hash value: 8645535281

---------------------------------------------------------------------------------------------
| Id  | Operation                    | Name         | Rows  | Bytes | Cost (%CPU)| Time     |
---------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |              |   500K|  9765K|  1508K  (1)| 05:01:47 |
|   1 |  NESTED LOOPS                |              |       |       |            |          |
|   2 |   NESTED LOOPS               |              |   500K|  9765K|  1508K  (1)| 05:01:47 |
|   3 |    TABLE ACCESS FULL         | T1           |  1000K|  9765K|  8279   (1)| 00:01:40 |
|*  4 |    INDEX UNIQUE SCAN         | SYS_C0011891 |     1 |       |     1   (0)| 00:00:01 |
|   5 |   TABLE ACCESS BY INDEX ROWID| T2           |     1 |    10 |     2   (0)| 00:00:01 |
---------------------------------------------------------------------------------------------

By: Charles Hooper

Charles Hooper — Sat, 15 Jan 2011 23:15:37 +0000

Coskan,
I am hoping that the author of chapter 16 can explain what is happening. Strange things happen when you tell the optimizer that you want FAST PERFORMANCE – see my reply to Martin Preiss below.