duck ! brickbats en route !
"Ed Avis" <ed_at_membled.com> wrote in message
news:l1smhmj0cx.fsf_at_budvar.future-i.net...
> "Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> writes:
>
> >To date, every example I have seen of Oracle failing to handle
> >serialization properly has required the construction of an incorrect
> >data model, so I haven't been able to get excited about the issue.
>
> To me, this doesn't make much sense. If the database fails to handle
> transaction ordering correctly (according to the database-theory
> definition of 'serializable'), then it is not much consolation to wave
> hands and say this is an incorrect data model. It's a bad data model
> only because it's a case that the database gets wrong!
>
> I suppose it would be okay if on executing the SQL, the programmer saw
> a warning:
>
> This transaction cannot be guaranteed to execute in a serializable
> order, even though you have chosen the 'serializable' isolation
> level. Please see the documentation for more details.
>
> but it isn't like that, the database is quietly choosing to misorder
> the operations in two transactions, and that's something I would
> expect more from MySQL than an industrial-strength RDBMS like Oracle,
> which is good at maintaining transaction isolation in other cases.
>
> --
> Ed Avis <ed_at_membled.com>
Received on Sun Feb 08 2004 - 11:22:44 CST
