Comments on: Multipart cursor subexecution and PRECOMPUTE_SUBQUERY hint http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/ Oracle troubleshooting, internals and performance tuning Mon, 15 Mar 2010 01:50:11 +0000 http://wordpress.org/?v=2.9.2 hourly 1 By: dalin qin http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/comment-page-1/#comment-1378 dalin qin Thu, 09 Apr 2009 18:06:38 +0000 http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/#comment-1378 wonderful article! wonderful article!

]]> By: Tanel Poder http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/comment-page-1/#comment-934 Tanel Poder Thu, 05 Feb 2009 04:56:54 +0000 http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/#comment-934 @Karl In 11g if you have the Diag and Tuning pack licenses, you can use V$SQL_MONITOR (or OEM for that). Greg Rahn has written about it: http://structureddata.org/2008/01/06/oracle-11g-real-time-sql-monitoring-using-dbms_sqltunereport_sql_monitor/ In case you don't have Diag/Tuning pack licenses or you're not on 10g, then you can use a technique of getting stack trace of the process and translating the function names to execution plan rowsource names. It's not as good as thorough solution as the 11g SQL monitoring thing though.. http://blog.tanelpoder.com/2008/06/15/advanced-oracle-troubleshooting-guide-part-6-understanding-oracle-execution-plans-with-os_explain/ @Karl

In 11g if you have the Diag and Tuning pack licenses, you can use V$SQL_MONITOR (or OEM for that).

Greg Rahn has written about it:
http://structureddata.org/2008/01/06/oracle-11g-real-time-sql-monitoring-using-dbms_sqltunereport_sql_monitor/

In case you don’t have Diag/Tuning pack licenses or you’re not on 10g, then you can use a technique of getting stack trace of the process and translating the function names to execution plan rowsource names. It’s not as good as thorough solution as the 11g SQL monitoring thing though..

http://blog.tanelpoder.com/2008/06/15/advanced-oracle-troubleshooting-guide-part-6-understanding-oracle-execution-plans-with-os_explain/

]]> By: Tanel Poder http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/comment-page-1/#comment-933 Tanel Poder Thu, 05 Feb 2009 04:54:10 +0000 http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/#comment-933 @Gary, When you execute a query, Oracle takes the current system SCN and runs the query (consistent gets) as of that SCN. It's possible to pass a requested SCN to another query as well (like the SELECT ... FROM t AS OF SCN 123 for example), so Oracle is probably doing something similar. It gets the SCN for the subquery precomputation statement and then passes the same SCN on to the main query. That way there would be no consistency issue. I haven't tested this, if the kernel developer has forgotten to do this, then it's a bug.. @Gary,

When you execute a query, Oracle takes the current system SCN and runs the query (consistent gets) as of that SCN. It’s possible to pass a requested SCN to another query as well (like the SELECT … FROM t AS OF SCN 123 for example), so Oracle is probably doing something similar. It gets the SCN for the subquery precomputation statement and then passes the same SCN on to the main query. That way there would be no consistency issue. I haven’t tested this, if the kernel developer has forgotten to do this, then it’s a bug..

]]> By: Tanel Poder http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/comment-page-1/#comment-928 Tanel Poder Wed, 04 Feb 2009 23:23:58 +0000 http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/#comment-928 Hi Mladen, Also the join itself will be replaced with a filter operation, so there will be no joining of multiple row sources, just filtering one. My (very) brief lab tests didn't show any performance improvement as CPU usage is considered though. As I wrote in the end of the post, I intended this post only for demonstrating some new capabilities of Oracle (and intro some of my scripts), but I haven't seen any tuning/performance value for non-OLAP queries yet. .. David, Hmm, interesting idea. Btw, Oracle 11.1.0.7 should be able to push both filter conditions and also join's bloom filters down to cell level, so join conditions can be "exadata optimized" even without this subquery precomputation thing (and it's possible to convert IN subqueries to semijoins..) Hi Mladen,

Also the join itself will be replaced with a filter operation, so there will be no joining of multiple row sources, just filtering one. My (very) brief lab tests didn’t show any performance improvement as CPU usage is considered though.

As I wrote in the end of the post, I intended this post only for demonstrating some new capabilities of Oracle (and intro some of my scripts), but I haven’t seen any tuning/performance value for non-OLAP queries yet.

..

David,

Hmm, interesting idea.

Btw, Oracle 11.1.0.7 should be able to push both filter conditions and also join’s bloom filters down to cell level, so join conditions can be “exadata optimized” even without this subquery precomputation thing (and it’s possible to convert IN subqueries to semijoins..)

]]> By: Mladen Gogala http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/comment-page-1/#comment-927 Mladen Gogala Wed, 04 Feb 2009 10:46:45 +0000 http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/#comment-927 This is a very neat hint but I don't see it as being a particularly useful one. The cost of accessing the "list of values" table is simply transferred to the parse phase, instead of the execution phase. Do you see it as being a useful optimization tool? If so, can you expand? Kindest regards, This is a very neat hint but I don’t see it as being a particularly useful one. The cost of accessing the “list of values” table is simply transferred to the parse phase, instead of the execution phase. Do you see it as being a useful optimization tool? If so, can you expand?
Kindest regards,

]]> By: David Aldridge http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/comment-page-1/#comment-922 David Aldridge Mon, 02 Feb 2009 21:14:44 +0000 http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/#comment-922 I would expect this hint to be pretty handy (internally) for queries against fact tables on the Exadata storage platform also. It's a nice complement to a star transformation. I would expect this hint to be pretty handy (internally) for queries against fact tables on the Exadata storage platform also. It’s a nice complement to a star transformation.

]]> By: Log Buffer #133: A Carnival of the Vanities for DBAs http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/comment-page-1/#comment-891 Log Buffer #133: A Carnival of the Vanities for DBAs Fri, 30 Jan 2009 17:17:25 +0000 http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/#comment-891 [...] Tanel Poder gets in-depth with cursors in Oracle. He calls it Multipart cursor subexection and PRECOMPUTE_SUBQUERY hint. [...] [...] Tanel Poder gets in-depth with cursors in Oracle. He calls it Multipart cursor subexection and PRECOMPUTE_SUBQUERY hint. [...]

]]> By: Gary http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/comment-page-1/#comment-877 Gary Thu, 29 Jan 2009 00:45:49 +0000 http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/#comment-877 Nice internals piece. My initial thought is that, if the subquery is broken out and executed as a separate SQL in its own right, then it may no longer operate as of the same SCN. A change could, theoretically, be committed between the two such that the query results aren't accurate at a single point in time. Nice internals piece.
My initial thought is that, if the subquery is broken out and executed as a separate SQL in its own right, then it may no longer operate as of the same SCN. A change could, theoretically, be committed between the two such that the query results aren’t accurate at a single point in time.

]]> By: Tanel Poder http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/comment-page-1/#comment-868 Tanel Poder Tue, 27 Jan 2009 21:10:39 +0000 http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/#comment-868 Hi Christo, As it's undocumented and (I think) it's also meant for OLAP option queries, I don't recommend anyone actually using this for trying to tune their normal queries. This article aims to explain yet another layer of complexity in process of interpreting SQL and converting it into execution plan(s). Hi Christo,

As it’s undocumented and (I think) it’s also meant for OLAP option queries, I don’t recommend anyone actually using this for trying to tune their normal queries.

This article aims to explain yet another layer of complexity in process of interpreting SQL and converting it into execution plan(s).

]]> By: Christo Kutrovsky http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/comment-page-1/#comment-866 Christo Kutrovsky Tue, 27 Jan 2009 15:28:24 +0000 http://blog.tanelpoder.com/2009/01/23/multipart-cursor-subexecution-and-precompute_subquery-hint/#comment-866 That's a pretty cool feature. I wonder now what's a good practical case where this will come handy. That’s a pretty cool feature.

I wonder now what’s a good practical case where this will come handy.

]]>