Comments on: Join Surprise

By: Log Buffer #210, A Carnival of the Vanities for DBAs

Log Buffer #210, A Carnival of the Vanities for DBAs — Fri, 15 Feb 2013 13:20:26 +0000

[...] Lewis was surprised by a join. If he was surprised, I’m sure everyone else will be too. Did you know that the order of join [...]

By: Jonathan Lewis

Jonathan Lewis — Tue, 28 Dec 2010 15:22:32 +0000

Pavel,

Thanks for checking.

By: Pavel Ermakov

Pavel Ermakov — Tue, 28 Dec 2010 10:12:37 +0000

Reproduces in 10.2.0.5.

By: Jonathan Lewis

Jonathan Lewis — Tue, 21 Dec 2010 08:09:37 +0000

Srivenu,

Interesting, and worrying, case – thanks for posting it.

I’ll have to take a closer look at the implications. Your example is about composite partitioned tables and parallel execution, but I wonder if it would generalise to more cases of multi-column hash joins.

By: ksrivenu

ksrivenu — Mon, 20 Dec 2010 16:53:21 +0000

Jonathan,
Some cases i had seen it to make a difference is during joins involving composite partitioned tables.
http://srivenukadiyala.wordpress.com/2010/08/09/plan_hash_value-limitations-pq_distribute-skew-in-data-distribution-across-parallel-slaves/
regards
srivenu

By: Jonathan Lewis

Jonathan Lewis — Fri, 17 Dec 2010 08:52:09 +0000

Vyacheslav,

Thanks for running the test and passing on the result.

By: Vyacheslav Rasskazov

Vyacheslav Rasskazov — Thu, 16 Dec 2010 23:14:29 +0000

Reproduces on 11.2.0.2.

By: Jonathan Lewis

Jonathan Lewis — Thu, 16 Dec 2010 19:05:49 +0000

Flado,

Thanks for that; and thanks for demonstrating one of the benefits of test cases – you can reproduce them with variations and different versions of Oracle to test their boundaries.

Anyone for 11.2 ?

By: Flado

Flado — Thu, 16 Dec 2010 08:19:25 +0000

I don’t have an 11g database at hand, but the behaviour reproduces on 10.2.0.3 as well. And it doesn’t need to be an outer join – an inner join behaves just the same.