Thanks for the update.
That’s the problem with publishing comments on undocumented mechanisms – Oracle changes the mechanisms.
I’ve also added a footnote to the article.

I have noticed that following query returns empty attr1 in my DB version 11.2.0.3.
attr5 contains actual hint. Just for information.

select rat.attr1
from
        wri$_adv_tasks     tsk,
        wri$_adv_rationale rat
where
        tsk.name = {sql tuning task name}
and     rat.task_id = tsk.id ;

SQL> select rat.attr1
  2  from
  3          sys.wri$_adv_tasks     tsk,
  4          sys.wri$_adv_rationale rat
  5  where
  6          tsk.name = 'TASK_1291'
  7  and     rat.task_id = tsk.id ;

ATTR1
-----------------------------------------------------------------------------------------------------------------------------------------------------------


























25 rows selected.

Elapsed: 00:00:00.04
SQL> 1 select rat.attr5
SQL> r
  1  select rat.attr5
  2  from
  3          sys.wri$_adv_tasks     tsk,
  4          sys.wri$_adv_rationale rat
  5  where
  6          tsk.name = 'TASK_1291'
  7* and     rat.task_id = tsk.id

ATTR5
-----------------------------------------------------------------------------------------------------------------------------------------------------------

OPT_ESTIMATE(@"SEL$1", TABLE, "STREAMS$_APPLY_SPILL_MSGS_PART"@"SEL$1", SCALE_ROWS=0.01028336274)
OPT_ESTIMATE(@"SEL$1", TABLE, "STREAMS$_APPLY_SPILL_MSGS_PART"@"SEL$1", SCALE_ROWS=0.01028336274)
OPT_ESTIMATE(@"SEL$1", INDEX_SCAN, "STREAMS$_APPLY_SPILL_MSGS_PART"@"SEL$1", "I_STREAMS_APPLY_SPILL_MSGS_PT1", SCALE_ROWS=0.0005412296179)
OPT_ESTIMATE(@"SEL$1", INDEX_SCAN, "STREAMS$_APPLY_SPILL_MSGS_PART"@"SEL$1", "I_STREAMS_APPLY_SPILL_MSGS_PT1", SCALE_ROWS=0.0005412296179)
OPT_ESTIMATE(@"SEL$1", INDEX_SKIP_SCAN, "STREAMS$_APPLY_SPILL_MSGS_PART"@"SEL$1", "I_STREAMS_APPLY_SPILL_MSGS_PT1", SCALE_ROWS=0.01028336274)
OPT_ESTIMATE(@"SEL$1", INDEX_SKIP_SCAN, "STREAMS$_APPLY_SPILL_MSGS_PART"@"SEL$1", "I_STREAMS_APPLY_SPILL_MSGS_PT1", SCALE_ROWS=0.01028336274)
COLUMN_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", "FLAGS", scale, length=2)
COLUMN_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", "FLAGS", scale, length=2)
COLUMN_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", "FLAGS2", scale, length=1)
COLUMN_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", "FLAGS2", scale, length=1)
COLUMN_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", "SEQUENCE", scale, length=3 distinct=10000 nulls=0 min=1 max=10000)
COLUMN_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", "SEQUENCE", scale, length=3 distinct=10000 nulls=0 min=1 max=10000)
COLUMN_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", "DESTQUEUE", scale, length=0)
COLUMN_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", "DESTQUEUE", scale, length=0)
COLUMN_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", "MESSAGE", scale, length=913)
COLUMN_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", "MESSAGE", scale, length=913)
COLUMN_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", "TXNKEY", scale, length=2 distinct=19 nulls=0 min=2 max=20)
COLUMN_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", "TXNKEY", scale, length=2 distinct=19 nulls=0 min=2 max=20)
TABLE_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", scale, blocks=28954 rows=182916.895)
TABLE_STATS("SYS"."STREAMS$_APPLY_SPILL_MSGS_PART", scale, blocks=28954 rows=182916.895)
IGNORE_OPTIM_EMBEDDED_HINTS
IGNORE_OPTIM_EMBEDDED_HINTS
OPTIMIZER_FEATURES_ENABLE('8.1.7')
OPTIMIZER_FEATURES_ENABLE('8.1.7')

25 rows selected.

Elapsed: 00:00:00.14
SQL> select banner from v$version;

BANNER
--------------------------------------------------------------------------------
Oracle Database 11g Enterprise Edition Release 11.2.0.3.0 - 64bit Production
PL/SQL Release 11.2.0.3.0 - Production
CORE    11.2.0.3.0      Production
TNS for Solaris: Version 11.2.0.3.0 - Production
NLSRTL Version 11.2.0.3.0 - Production

Elapsed: 00:00:00.62

If you enable sql_trace (or 10046 at level 4 in particular) you will see that for each index Oracle checks the number of bytes in dba_segments. This will always return null (which is nvl()ed to zero) for the no_segment indexes. It is probably this check that makes the tuning task ignore such indexes.

Bear in mind, the tuning task code knows about no_segment indexes, because one of its internal mechanism is to “create” such indexes anyway.

I have been performing some tuning using dbms_sqltune. Several recommendations have come back for the use of a SQL Profile as well as additional indexes. As some of the tables are rather large I was trying to create the suggested indexes using the nosegment parameter and specifying the _use_nosegment_indexes session level parameter and manually running dbms_sqltune again. Unfortunately all attempts fail to pick up the nosegment indexes. Is this normal or is there a way I can get dbms_sqltune to consider the nosegment index? I have used dbms_stats.gather_index_stats but this has not helped.

Your response would be much appreciated.

Thanks for the pingback, and the note.

SQL Profiles and Outlines are ways of addressing two different problems – the profiles help if the data ratios are going stay constant of time, regardless of data volume; the outlines help if a specific execution path is always going to be a good idea as time passes, regardless of how the data ratios change. So each has its purpose.

I like your idea, though, of treating both as a temporary measure, and using an SQL profile as a quick way of getting to an outline that keep you going for a while.

In passing, outlines are deprecated in 11g – presumably because the technology is just a subset of the SQL Plan Management (SPM) technology.

Great info as always. One of the things that I have noticed is that sql profiles can and do “sour” over time (as mentioned by Alex). Since they do most of their work by applying scaling factors via the opt_estimate hint, when things change (statistics for example) the scaling factors can end up being way off. I’ve had some success using a profile to get the optimizer to do what it should and then creating an outline for the statement to lock it (while I try to figure out why it went wrong in the first place).

Also, I have been experimenting with setting up a blog and we have installed wordpress on one of our servers to play with. WordPress is pretty smart and it automatically created the pingpack entry above. I didn’t know it would do that – pretty cool! Although it was totally unintentional as we aren’t even sure that’s where it will live yet. I guess everything really is easy if you don’t know what you’re doing. Anyway, keep up the great work.