Re: Why is undo backed with redo for GTT

Richard,

> I know where you're coming from but the point is that as it causes changes
> to the *undo* tablespace, Oracle requires the redo for these changes to
> ensure the associates files can be recovered to a consistent state. But I
> agree it's all somewhat fundamentally unnecessary.

Actually I was just telling the opposite about rolling forward - the undo blocks for "temp" operations never should be reconstructed to consistent state after a crash - they won't ever get read (at least no information is obtained from them), instead overwritten.

> Me, if I was designing how GTT handled undo, I would implement a
> "deferred-like" undo segment(s) for the GTT that is stored in the temp
> tablespace so that associated undo would not require the generation of
redo.
>
> That said, I can immediately think of half a dozen reasons why this isn't
> practical, explaining why I'm not an Oracle developer ...

That was a good one :D

Tanel. Received on Thu Mar 06 2003 - 07:42:10 CST