RE: dbms_stats.auto_invalidate

> in a lot of plans. IIRC, auto_invalidate will invalidate 
> plans gradually over time, so it is not a decision of whether 
> to invalidate a plan or not but a decision which ones to 
> invalidate sooner and which ones later.
> But as I said, I could be totally wrong in that.

Yep, this is dependent on _optimizer_invalidation_period parameter, which is
set in seconds. 
It defaults to 18000, meaning that all "old" cursors should be invalidated
during 5 hour period since stats gathering.

This should avoid the "hard parse storm" when lots of cursors are dependent
on the analyzed objects...



Other related posts: