Re: How to reproduce an ORA-7445?

  • From: Andrey Kriushin <Andrey.Kriushin@xxxxxxxx>
  • To: Rich.Jesse@xxxxxx
  • Date: Sun, 31 Dec 2006 14:03:26 +0300

Sorry for late joining and with a little bias to the NLS conversions themselves.


Are you using NCHAR/NVARCHAR2/NCLOB types in your binds?
We had some issues with NCLOB in 9.2.0.5 or 6 (don't remember exact version). The diagnostics was ORA-00600 with no arguments.

What was discovered while working on SR is that when converting from NCLOB (or probably NVARCHAR) to main characterset of database, the server process had simply (i.e. without any latching/locking) overwritten some areas in the SGA, where NLS structures were stored, with the values pertinent to national characterset. Of course, these structures were restored to their original values when the conversion was completed. But in the middle of the conversion the other sessions sometimes got that ORA-600 without arguments.

Thanks to Kyle Hailey for his ksuse.c. This approach allowed us to catch those memory overwrites and construct the "reproducible test case".

HTH
--Andrey

Jesse, Rich wrote:
Hey all,

In our migration of a new ERP system in 10.2.0.2.0 on AIX, a user hit an
ORA-7445 (with no parameters!).  I file an SR, upload the resulting
trace file, troubleshooting happens, and it's determined we've hit a BUG
due to issues with NLS translation of dates in binds.  I'm then told
that unless we can reproduce the problem, there will be no patch.


--
//www.freelists.org/webpage/oracle-l


Other related posts: