Re: Solution for 01555

"Paul Brewer" <paul_at_paul.brewers.org.uk> wrote:
> <ctcgag_at_hotmail.com> wrote in message
> > I don't think Vlad is asking Oracle to return inconsistent data in lieu
> > of 1555s, but rather asking it to realize what's going on and take
> > appropriate actions to prevent the need for 1555s in the first place.
> >
> > I don't know whether he thinks this can be done without a major
> > performance impact, or if he has such a puritanical view that he
> > doesn't care about the impact.
> >
> >
> Fair enough, on most counts. But for how long does Vlad feel that Oracle
> should retain the old data on transactions committed after his query has
> started?
> I fear we are back to square one..."until all queries needing these data
> have finished".

I don't wish to speak for Vlad, but I think it would be until all queries (in read-committed mode) and/or transactions (in serializable or read-only mode) starting before the commit are done. Obviously this means that Oracle needs to keep track of the start time of all running queries/transactions, so we are now back at square two.

If I were an Oracle developer (or I guess an Oracle competitor) who wanted to make Vlad semi-happy, I think I'd go about it by adding another mode, "SET LONG RUNNING TRANSACTION ISOLATION LEVEL ....", then requiring that rollback data newer than the oldest LONG RUNNING transaction cannot be overwritten (except in dire straights, i.e. cannot extend error is emminent.)

Cheers,

Xho

-- 
-------------------- http://NewsReader.Com/ --------------------
Usenet Newsgroup Service              New Rate! $9.95/Month 50GB