On Thu, 11 Mar 2004, Bruno Wolff III wrote:
> On Thu, Mar 11, 2004 at 14:21:09 -0500,
> Robert Treat <xzilla@users.sourceforge.net> wrote:
> > On Thu, 2004-03-11 at 13:59, Peter Eisentraut wrote:
> > > Josh Berkus wrote:
> > > > > Isolation Levels
> > > > > (Support all four ANSI isolation levels (UR, CS, RS, RR).)
> > > >
> > > > I'm not sure about this one; I suspect that we do, however, since
> > > > MVCC, invented for the Postgres Project, has become a standard for
> > > > transaction isolation in the database industry.
> > >
> > > 7.5 will support all four isolation levels.
> > >
> >
> > I wasn't aware of anyone doing work on this... do you have any more
> > info?
>
> My memory of past discussion on this is that you would be able to set all
> 4 transaction isolation levels in a command, but you would really get
> the next level up for the two that we don't provide. This is OK since
> since they get at least the transaction isolation safety they need.
> The reason for doing this is portability.
I thought the other two were considered meaningless within the context of
MVCC... I.e. they were defined when row locking was the standard way of
storing data, and reflect that, and MVCC changed the rules so much that
the other two, which were a cheaper to implement locking model in a row
locking database, became no cheaper, and therefore there was no reason go
program a way to expose the data the way that locking model defined, since
doing so would provide poorer data integrity while actually being the same
speed or slower.
Or something like that.
So, are those isolation levels being actually implemented, or are we just
pretending to set the level in 7.5 but still using the next level higher?
