Re: UNDO: 10g-style behaviour in 18.104.22.168?
Date: Mon, 13 Apr 2009 12:51:43 -0700 (PDT)
On Apr 13, 10:16 am, "Jonathan Lewis" <jonat..._at_jlcomp.demon.co.uk> wrote:
> Thanks for supplying a couple of references.
> My comment was specifically in response to this point, though:
> "In the AutoExtend ON case, the value you set for "undo_retention" is
> the floor value but Oracle automatically adjusts the actual
> undo_retention it uses on the basis of query durations"
> i.e. the case when AutoExtend is ON - and my point was that I had seen
> the tuned_undoretention value in v$undostat drop BELOW the setting
> for parameter undo_retention with autoextend ON.
apologies, I got that wrong and thought it was about the "autoextend OFF" case. Your case seems to be odd, yes, and I would tend more to the "implementation bug" or "implementation limitation" explanation (given the "all available space tied to one large undo segment" information).
By the way, I thought that I've seen TUNED_UNDORETENTION in V$UNDOSTAT larger than UNDO_RETENTION with AUTOEXTEND = ON with long running queries, but I'm not entirely sure and therefore it might be wrong.
Oracle related stuff blog: