RE: Tracing FGAC/VPD differences between 9i and 10g

I apologize about running a one-side conversation here, but....

From what I can tell, the documentation is a bit screwed up (would not be the first time):
http://download-east.oracle.com/docs/cd/B19306_01/network.102/b14266/apd vcntx.htm#sthref2431

I am most concerned about context_sensitive and shared_context_sensitive.

For shared_context_sensitive, table 15-3 seems to indicate that the policy function only executes the first time the object is referenced in a session. We found that the policy only executed for the first object referenced. Hence, if you reference two objects, the first one gets cached (I have a working example for those that are curious). Does not the documentation imply that each object should have its own predicate?

> _____________________________________________
> From: Schultz, Charles
> Sent: Wednesday, April 26, 2006 8:24 AM
> To: 'Oracle-L'
> Subject: RE: Tracing FGAC/VPD differences between 9i and 10g
>
> Correction: Shared_context_sensitive does help in some situations, but
> in my "simple" example, we still have the same issue due to caching
> (with shared_context_sensitive).
>
> _____________________________________________
> From: Schultz, Charles
> Sent: Wednesday, April 26, 2006 8:20 AM
> To: 'Oracle-L'
> Subject: RE: Tracing FGAC/VPD differences between 9i and 10g
>
> Not having any prior experience with VPD, I am kinda jumping in the
> water on this one. We have made progress with various tracing options
> and now have other issues. Specifically, we would like to reduce the
> library cache latch contention due to heavy parsing caused by a policy
> type of "context_sensitive". "Shared_context_sensitive" seems to help,
> but "static" would be even better, if we can figure out how to
> appropriately deploy it.
>
> Currently, our predicate function revolves around campus code (as we
> are a multi-campus educational facility), hence the VPD tables each
> have a vpdi column for use with the campus code. The problem is that
> the table name is part of the column name (a "naming standard" from
> the 3rd party vendor), hence if we use a static policy, subsequent
> queries against VPD tables fail because, obviously, the function is
> cached with the first table name that is executed.
>
> My question for the list: what is the best compromise we can achieve?
> Granted, there are other VPD issues as the situation is a bit complex
> - I am starting simple since that is all I understand at the moment.
> =) I have been trying to read up on the documentation, but it tends to
> be distracting when people keep asking questions as if I know the
> answer. *grin*
>
> _____________________________________________
> From: Schultz, Charles
> Sent: Tuesday, April 25, 2006 12:31 PM
> To: Oracle-L
> Subject: Tracing FGAC/VPD differences between 9i and 10g
>
> Does the sql trace facility (ie, event 10046) in 10g do better
> recursive tracing than 9i? From some tests we are running, we are
> seeing sql statements under 10g that do not show up under 9i (same
> application). I tried to scour the Concepts guide, but did not find
> anything relevant there (perhaps I missed it?).
>
> TIA,
>
> charles schultz
> oracle dba
> aits - adsd
> university of illinois
>
>

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