Comments on: SAS Bug

By: Log Buffer #125: a Carnival of the Vanities for DBAs

Log Buffer #125: a Carnival of the Vanities for DBAs — Fri, 28 Nov 2008 17:46:39 +0000

[...] “I see that Tom Kyte has found a nasty little bug waiting to trap a few unlucky people as they patch to 10.2.0.4, or upgrade to 11g.” So writes Jonathan Lewis in his item on the SAS bug. [...]

By: Jonathan Lewis

Jonathan Lewis — Thu, 27 Nov 2008 21:54:17 +0000

Randolf,

Thanks for the link and the references to Metalink. I’ve just run my test case against 10.2.0.3 – and it does, indeed, display the same problem, with the same workaround.

I guess your article and the bug notes also explain the anomaly that Scott reported.

By: Randolf Geist

Randolf Geist — Thu, 27 Nov 2008 20:58:26 +0000

Jonathan,

this nasty bug actually already appeared in the 10.2.0.3 patch set and was meant as a “bug fix”, see Metalink note 452466.1 and corresponding bug 6113096. I’ve written a blog note about this issue a couple of months ago:

http://oracle-randolf.blogspot.com/2008/02/nasty-bug-introduced-with-patch-set.html

Repeating your test case in the 11.1.0.7 patch set (32bit-Windows) it’s looks like the bug has been fixed although I cannot find it in the “list of bug fixes”.

Regards,
Randolf

By: Jonathan Lewis

Jonathan Lewis — Thu, 27 Nov 2008 20:30:12 +0000

Jeff, It's always good to think of different ways of extending the test. That was one I tried, and I also tried running a couple of queries before the one I printed in case this made the optimizer "notice" that the parameter had changed. lascoltodelvenerdi, I find that I discover all sorts of extra things by accident when designing a test for a specific problem. In this case (in 11g), I found that my usual trick of using the /*+ gather_plan_statistics */ hint followed by the extended call to dbms_xplan.display_cursor() randomly failed to find the cursor. (Cause and workaround still to be determined). Scott, Thanks for the note. I'll have to look into that when I get a bit of time.

By: Scott

Scott — Wed, 26 Nov 2008 17:13:18 +0000

Setting the sort_area_size twice only seems to work when the sort is done serially. It is still ignored when the sort is done in parallel.

Also if I “alter session force parallel dml” and set the 10032 trace on an insert as select then the both the query and the index maintenance become serial.

By: lascoltodelvenerdi

lascoltodelvenerdi — Wed, 26 Nov 2008 13:36:44 +0000

This "bug" remember me a similar case with External tables. Sometime if you issue an "alter table" because you want to add a file, you must do it twice, or the file will not be added. Rather strange...no? (Must say, that the set up that shows this particular thing, shows others, like you do a concatenation in a string to build a big dinamic-sql and you find that the end of the query is cut away. You rewrite the concatenation, something like a||b||c become ab||c, and everything works well.)

By: Jeff Moss

Jeff Moss — Wed, 26 Nov 2008 09:02:11 +0000

Is it possible that it’s not that you need to run the statement twice to work around it…but just wait long enough…i.e. is the first ALTER SESSION statement taking effect after a period of time or after you issue it for the second time?

…I know, I should test that…but no free time currently.