Re: SQL High version count because of too many varchar2 columns
Date: Fri, 14 Sep 2012 13:10:37 -0700 (PDT)
Message-ID: <1347653437.75904.YahooMailClassic_at_web184803.mail.gq1.yahoo.com>
> We have tried 10503 event in QA environment, but it got ORA-01483 errors.
> And I searched for this event, seems it's buggy and doesn't work as
> expected in all versions.
It's interesting that this note
CDC or Streams Apply Process Returns Error ORA-1483 "Invalid length for DATE
or NUMBER bind" [ID 759642.1]
suggests setting event 10503 as the solution for ORA-1483.
Indeed event 10503 sounds good but doesn't work well. You can only set it with an instance bounce after you set it in pfile or spfile. You can check it's set with "oradebug eventdump" or check PGA variable kxsusrgl (oradebug dumpvar pga kxsusrgl). Bug 10274265 says "we set the user defined graduated length (kxsusrgl) when we initialize the UGA, ie. the variable is not set if we set event 10503 in the session". Other than UGA/PGA mixup, that statement is confusing in that the variable *is* set when you set event 10503.
By the way, what is your session_cached_cursors? Could setting it to 0 help?
If everything fails, I think it's better to have the developers artificially introduce random strings (maybe just a few) to counteract the "benefit" of bind variables.
Yong Huang
-- http://www.freelists.org/webpage/oracle-lReceived on Fri Sep 14 2012 - 15:10:37 CDT