Comments on: Rule Rules

By: Israel

Israel — Tue, 14 May 2013 20:18:41 +0000

Did you got any resolution on this issue from Oracle?

By: Log Buffer #185, a Carnival of the Vanities for DBAs

Log Buffer #185, a Carnival of the Vanities for DBAs — Wed, 13 Feb 2013 16:39:00 +0000

[...] also shows how Oracle uses the /*+ rule */ hint in some internal views – and everyone knows you should never do [...]

By: Owen Ireland

Owen Ireland — Thu, 15 Apr 2010 16:40:29 +0000

As the person who Jonathan spoke to a few weeks ago I’d just like to point out a few things.

1) Although v$datafile itself doesn’t have the rule hint in it, it is itself a view on gv$datafile so you still get a rule based plan.

SQL> select view_definition from v$fixed_view_definition where view_name like ‘V$DATAFILE’;

VIEW_DEFINITION
——————————————————————————–
select … from GV$DATAFILE where inst_id = USERENV(‘Instance’)

(columns removed for brevity)

2) The patch for bug 5251842 isn’t in the PSUs for the non-Windows platforms. On our platform Linux x86-64 it’s not even included in 10.2.0.4.3. It is available as a one-off patch though, so many thanks to Michael for that.

3) The all_rows hint makes the difference between 3 seconds and 3 hours on our largest database which has 11,000 datafiles !

3) I am sure this bug is to blame for our standby database taking hours to decide which logs to obtain when performing gap resolution. It must internally use v$archive_gap. When I apply the patch for bug 5251842 I’ll let you know.

4) My E-mail was blogged with my full permission to Jonathan in the interests of making the Oracle community aware of this issue.

By: Timur Akhmadeev

Timur Akhmadeev — Thu, 15 Apr 2010 08:59:17 +0000

Oracle can include base step in recursion which will stop it. I mean, incorporate RBO to CBO for data dictionary queries. Since it hasn’t been done for years, I suspect it isn’t an easy task.

By: Marcus Moennig

Marcus Moennig — Wed, 14 Apr 2010 20:27:34 +0000

Jonathan,

that makes very much sense. And it puts the “deprecated” status of the RBO into perspective.

By: Timur Akhmadeev

Timur Akhmadeev — Wed, 14 Apr 2010 13:43:38 +0000

That's why bundled patches are recommended: >From 10.2.0.4 onwards each Windows bundled patch goes through full regression testing because it includes a large number of dependent fixes, CPU, DST and PSU. Oracle recommends all customers apply the latest bundle as soon as it is convenient irrespective of encountering a specific problem resolved by the bundle. This particular bug was fixed in P5.

By: Asif Momen

Asif Momen — Wed, 14 Apr 2010 12:09:24 +0000

Timur,

On a contrary, I see one view in my 10.2.0.4 (32-bit WinXP) database.

SQL> select * from v$version;

BANNER
—————————————————————-
Oracle Database 10g Enterprise Edition Release 10.2.0.4.0 – Prod
PL/SQL Release 10.2.0.4.0 – Production
CORE 10.2.0.4.0 Production
TNS for 32-bit Windows: Version 10.2.0.4.0 – Production
NLSRTL Version 10.2.0.4.0 – Production

SQL>
SQL>
SQL> select view_name from v$fixed_view_definition where lower(view_definition) like ‘%/*+%rule%*/%’;

VIEW_NAME
——————————
GV$DATAFILE

SQL>

By: Jonathan Lewis

Jonathan Lewis — Wed, 14 Apr 2010 10:53:48 +0000

Marcus,

That’s the problem of recursion: to optimize a statement you need to read data about the objects from the data dictionary – which means you have to query the data dictionary – which means you have to optimize the statements that query the data dictionary … but you can’t optimize the data dictionary queries until you’ve read the data dictionary information that defines the data dictionary.

By: Marcus Mönnig

Marcus Mönnig — Wed, 14 Apr 2010 10:29:01 +0000

The optmizer doesn’t seem to trust itself though… ;-)

select parsing_schema_name,sql_text from v$SQL where SQL_TEXT like ‘%/*+%rule%*/%’;

By: Jonathan Lewis

Jonathan Lewis — Tue, 13 Apr 2010 20:37:14 +0000

Michael,

Thanks for the reference.